Find OpenCV Projects on Upwork with Vollna

We turn real human work into training data for robotics. We capture first-person footage of skilled workers and transform it into structured, quality-checked datasets that help robots learn to operate in the physical world.

We're hiring an exceptional, hands-on Python engineer to own the architecture behind that growth: how large volumes of footage enter our platform, move between storage and compute, become usable training data and reach customers reliably.

Your mission is to build a platform capable of handling multiple terabytes of incoming footage per day and growing toward petabyte-scale storage—with rigorous quality controls and sustainable processing costs.

You'll work directly with the founders, take ownership of the existing platform and shape its architecture as we scale.

You will:
- Own the complete system architecture: connect capture devices and ingestion stations, cloud storage, processing workers, metadata services, quality checks and dataset delivery.
- Design reliable ingestion at scale: resumable uploads, integrity verification, deduplication, bandwidth management and recovery from interrupted transfers across many recording sites.
- Architect storage across Backblaze B2 and AWS: define landing storage, processing access, archival, retention and restoration workflows. Make deliberate decisions about durability, retrieval times, transfer costs and storage economics.
- Build distributed Python and GPU processing: design queues, scheduling, parallel execution, checkpointing, retries and backpressure so the platform handles growing workloads and recovers cleanly from partial failures.
- Own the video and ML pipeline: integrate and improve face blurring, privacy checks, hand-pose estimation, capture-quality assessment and dataset packaging.
- Protect data integrity throughout processing: preserve frames, timestamps and recording boundaries; track source files, model versions and processing history; make outputs reproducible and auditable.
- Measure and improve performance: identify bottlenecks across networking, storage, decoding, CPU and GPU workloads. Improve throughput and cost per processed footage-hour while meeting quality requirements.
- Build operational reliability: establish monitoring, actionable alerts, automated quality gates, deployment practices and recovery procedures that let a small team operate a large platform.

We're looking for someone with:
- Exceptional Python engineering skills, including concurrency, multiprocessing, profiling, memory management and debugging complex production systems.
- Proven architecture experience with substantial data volumes: systems they personally designed, shipped and operated, with concrete evidence of scale and reliability.
- Strong knowledge of AWS, object storage and distributed systems, including storage lifecycle design, queues, access controls, failure recovery and cloud cost optimisation.
- Experience designing workflows that remain correct through duplicate events, interrupted jobs, worker failures and retries.
- Practical experience with video processing, GPU workloads or production ML infrastructure.
- Strong database and metadata modelling skills, with an understanding of how to keep large media assets and their processing records consistent.
- The judgment to make sound architectural decisions, implement them personally and take responsibility for how the system behaves in production.

Experience with Backblaze B2, AWS S3 and archival storage, PyTorch, OpenCV, FFmpeg, Modal, robotics datasets or multi-camera recordings would be especially valuable.

Our environment includes Python, GPU inference, cloud object storage, a Go ingestion layer and a TypeScript control plane. You'll own the decisions that connect these components into a reliable platform and guide its evolution.

You'll join at a stage where your engineering decisions directly shape the company's capacity, data quality and economics. The role offers substantial technical ownership, close collaboration with the founders and room to grow into broader technical leadership as the team expands.

Windows app that works with a custom wired mouse.

Partner with a hardware dev (WIP)

Hardware/firmware is a separate contractor.
The app captures a region of the screen, finds a target, and sends small USB HID feature reports (Δx, Δy, confidence, flags, heartbeat) to the mouse. Firmware on the mouse mixes that with the real optical sensor. The PC app must not move the cursor. No PyAutoGUI, SendInput, mouse_event, interception, or virtual mouse.
v1 target finder is color
User picks a color (eyedropper) + tolerance / HSV. Scan the ROI only. Return position + confidence.
AI is a toggle, not the job
UI: Color | AI. One detect(frame) → x, y, confidence, source interface so YOLO/ONNX can be dropped in later. AI path can be “not implemented” in milestone 1.
Also in the app
Login (real or stub for milestone 1)
Subscription/entitlement gate (can mock): if logged out / expired, stop sending packets
Panel: mouse connected, enable, confidence floor, ROI, DPI field, color picker
Hotkey to kill assist
Overlay off by default
Stack
Python + PyQt/PySide or C++/Qt. Capture: DXGI preferred; mss/bettercam OK for v1. Color: OpenCV fine. HID: hidapi. One language is enough.
Milestone 1 (quote this fixed)
Runnable Windows exe:
Login or login stub + main panel
ROI capture preview
Color detect on a test shape (e.g. colored circle on screen)
Color | AI toggle (AI stub)
Dummy HID send to a mock or our bench mouse
Quit / disable stops packets
Short README
Out of scope
PCB, firmware, wireless, training YOLO, host-side aiming.

Skills: Windows Desktop, Python or C++, Qt or PyQt, OpenCV, API Integration, USB.

I am looking for a technical consultation for an academic project that combines mmWave radar, Raspberry Pi, computer vision, and a Pan-Tilt camera for real-time object detection, tracking, and classification. I need advice on hardware selection, system design, integration, and implementation.

Titre du projet: Application d'identification portes et fenêtres

Description: Dans le domaine de la construction, de la rénovation et de la gestion immobilière, l’identification précise des éléments architecturaux tels que les portes et les fenêtres est essentielle. Actuellement, cette tâche repose souvent sur des mesures manuelles et des inspections visuelles, ce qui peut être chronophage et sujet à des erreurs. L’avènement de la vision par ordinateur et des technologies mobiles offre une opportunité unique d’automatiser ce processus à l’aide d’une simple photo prise avec un smartphone.

Objectifs:
Développer une application Android/Iphone capable de :
- Capturer une image d’une porte ou d’une fenêtre à l’aide de l’appareil photo du téléphone.
- Identifier automatiquement le type d’ouverture (porte simple, porte double, fenêtre coulissante, fenêtre à battants, etc.).
- Estimer les dimensions réelles (hauteur, largeur) de l’objet détecté à partir de l’image, en utilisant des techniques de calibration ou des objets de référence.
- Générer un rapport contenant les informations extraites (type, dimensions, localisation GPS, date/heure).

Objectifs à atteindre:
- Application Android fonctionnelle.
- Documentation technique (architecture, choix technologiques, manuel utilisateur).
- Rapport de PFE détaillé.
- Présentation finale avec démonstration.

Expertises requises - technologies proposées :
- Langage : Kotlin ou Java pour Android.
- Frameworks : TensorFlow Lite ou MediaPipe pour la vision par ordinateur.
- Outils complémentaires : OpenCV, ARCore (pour la mesure), Firebase (pour le stockage), ML Kit (pour la classification).
- Base de données : SQLite ou Firebase Realtime Database.

Skills: Java, Graphic Design, Mobile App Development, iPhone, Android, Kotlin, OpenCV, ARCore

B2B Cold Caller / Appointment Setter – Vending Machine Placement – Los Angeles

I own a vending machine business in the Los Angeles / San Fernando Valley area and am looking for an experienced B2B cold caller to help identify qualified locations for vending machines.

You will be contacting businesses such as:

- Manufacturing facilities
- Warehouses and distribution centers
- Occupational health / medical facilities
- Large auto and industrial businesses
- Employee-heavy businesses
- Sports/recreation facilities
- Other businesses with significant employee or customer traffic

Your goal is NOT to close the sale.

Your job is to:

1. Call the business.
2. Get past the receptionist/gatekeeper.
3. Identify the person responsible for vending, facilities, employee breakroom services, purchasing, operations or property management.
4. Determine whether they currently have vending machines.
5. If they do, determine whether they are satisfied with their current service.
6. If they do not, determine whether they would consider free vending-machine placement.
7. Obtain the decision-maker's name, direct phone number and/or email.
8. Schedule a call or site visit with me when appropriate.
9. Record every result in a shared Google Sheet.

I will provide the initial leads, qualification criteria and script. I also want someone comfortable suggesting improvements to the script based on actual conversations.

Requirements

- Previous B2B cold-calling experience
- Experience speaking with U.S. businesses
- Strong, clear spoken English
- Confident getting past gatekeepers
- Comfortable calling manufacturing, industrial and medical businesses
- Available during Pacific Time business hours
- Reliable phone/internet setup
- Accurate call tracking

Spanish-speaking is a plus.


When applying, please send:

1. Your B2B cold-calling experience.
2. Industries you've called previously.
3. Your typical number of calls per hour/day.
4. Your typical appointment or qualified-lead rate.
5. A 60-second voice recording doing a mock cold call to a manufacturing facility offering free vending-machine placement.

Please do not apply without the voice recording.

We are seeking an experienced computer vision developer or team to build AI analytics using our existing CCTV cameras across Bahrain and Saudi Arabia.

Phase One covers kitchen mask, hairnet and apron compliance; cutting-board and knife color checks; staff identification while masked; and departmental alerts. Phase Two covers unattended customers, kitchen pass delays, delivery pickup delays and boutique replenishment.

Our priority is a reliable, scalable solution with low recurring costs, source-code ownership and full technical handover. Please review the attached scope and propose your approach, pilot plan, timeline and pricing.

The goal is to bundle four AI utilities into one intuitive Streamlit interface written entirely in Python.

1. Spam Detection
The core focus is robust spam prediction models. Accuracy, speed, and clear probability outputs matter more to me than classic rule-based email filters, but the module should be structured so other filtering approaches can be plugged in later.

2. Mushroom Classification
A computer-vision model (CNN or transfer-learning) should let users upload a mushroom photo and instantly see whether it is edible or poisonous, with confidence scores and a short explanation of key visual cues.

3. Text Summarization
I need concise, human-readable summaries of long articles. The user will paste or upload text up to ~5 000 words and receive a summary adjustable by ratio or target sentence count. If the same component can later handle documents or web pages, even better, but the long-article use case comes first.

4. Image Processing
Please wire in an OpenCV-based playground tab for common transformations (resize, grayscale, edge detection, simple filters). I am open to expanding this to detection or segmentation later, so keep the code modular.

Interface
A left-side navigation menu should switch between the four tasks. Each page must allow file or text input, display results instantly, and log key metrics in the sidebar. Clean, material-like styling is enough; no heavy frontend work required.

Deliverables
• Fully working Streamlit app with the four pages integrated
• All Python source files, requirements.txt, and a brief README explaining setup, model training, and how to extend each module
• Pre-trained weights or clear instructions to recreate them
• Short video or screenshot walkthrough confirming each feature operates as described

Acceptance Criteria
The app launches with a single `streamlit run` command, processes example inputs for every module without errors, and achieves respectable accuracy on publicly available test data (exact numbers fine-tuned during hand-over).

Tooling keywords for reference: TensorFlow or PyTorch, scikit-learn, Transformers, HuggingFace, NLTK, spaCy, OpenCV, Streamlit Components.

Timeline and milestones are flexible; code quality and clarity come first.

Skills: C Programming, Python, Software Architecture, Machine Learning (ML), Data Science, OpenCV, Computer Vision, Deep Learning, Streamlit, Convolutional Neural Network

Job highlights
8+ years experience in AI/GenAI with expertise in LLMs, RAG, Azure AI Foundry, LangChain, Agentic AI, Docker, Kubernetes

Design, build and deploy LLM, GenAI & Agentic AI solutions; lead multi-agent system architecture; implement enterprise-grade RAG solutions; optimize AI models; integrate with cloud and APIs; drive AI development using modern SDLC tools; ensure AI governance and ethics

8 - 13 Years

3 Vacancies

Salary: freelance monthly rolling contract (figures tbd)

Remote role

Must have key skills
AI Foundry,
Langchain,
LLM,
AIML,
Python

Other key skills
Tensorflow,
GenAI,
Artificial Intelligence,
Machine Learning,
Deep Learning,
Pytorch,
NLP,
Opencv,
Image Processing

Job description

What you’ll do
HIRING | 8yrs+ Lead AI Engineer GenAI & Agentic AI

Are you an experienced AI Engineer / GenAI Architect ready to build and lead next-generation LLM, RAG and Agentic AI solutions?

We are looking for Lead AI Engineers with 8+ years of experience, including strong hands-on experience in AI/GenAI, LLMs, RAG, Azure AI Foundry / Microsoft Foundry, and Agentic AI.

Experience: 8+ Years Role: Senior / Lead AI Engineer Domain: AI / GenAI / Agentic AI
Location: Remote role

Key Responsibilities
Design, build and deploy LLM, GenAI & Agentic AI solutions
Lead Multi-Agent System design and architecture
Architect and implement enterprise-grade RAG solutions
Leverage Azure AI Foundry / Microsoft Foundry for model orchestration, deployment, and evaluation
Optimize AI/GenAI/Agentic AI models for performance and scalability
Integrate AI solutions with Cloud, APIs and backend systems
Drive AI development using modern AI SDLC tools such as GitHub Copilot, Claude Code, Codex, etc.
Build and deploy solutions using CI/CD pipelines
Drive technical solutioning and collaborate with cross-functional teams
Contribute to AgentOps, AI governance, security, ethics and responsible AI

Mandatory Skills
8+ years of overall experience
Strong hands-on AI / GenAI experience
Azure AI Foundry / AI Foundry / Microsoft Foundry
Python
LangChain & LangGraph
LLMs & Generative AI
RAG Architecture
Agentic AI / Multi-Agent Systems
Docker & Kubernetes
CI/CD
Cloud & API integration
Enterprise AI Platforms

Experience with one or more:
Azure AI Foundry / Microsoft AI Foundry AWS Bedrock / AgentCore / AgentBricks Databricks Mosaic AI Gemini Enterprise Agent Platform / Agent Studio Snowflake Cortex Agents

Preferred Background
Candidates from Data Engineering, Data Science or Data-focused software engineering backgrounds are preferred.

2+ years of experience working on data projects will be an added advantage.

Soft Skills VERY IMPORTANT
Excellent communication skills are a must.

The ideal candidate should be able to:
Lead technical discussions • Communicate complex AI concepts clearly • Collaborate with engineering, data and business teams • Own end-to-end delivery • Mentor and guide technical teams

If you are an AI Engineer / GenAI Engineer / AI Architect / Lead AI Engineer with 8+ years of experience and strong expertise in RAG, LLMs, Azure AI Foundry & Agentic AI, we’d love to connect!

Already have a CCTV camera network installed on-site and are looking for a computer vision developer/team to build a real-time PPE (Personal Protective Equipment) compliance detection system that plugs into our existing camera feeds — no new hardware installation required.

The system should be able to:
- Detect workers in frame and classify them as SAFE / UNSAFE based on PPE worn (helmet, safety vest/jacket, hand gloves, safety harness/hook, etc.)
- Overlay bounding boxes with live labels on the video feed
Log unsafe events with a timestamped snapshot, camera ID, and missing-PPE detail (e.g., "Missing: helmet")
- Match against a registered worker database (face ID) for accountability
- Support an alarm/alert center for real-time unsafe notifications
- Support IN/OUT tracking at gates/turnstiles (worker headcount on site)
- Provide a simple dashboard for management to review live feed + logs + analytics

We've attached reference screenshots of a similar system

Skills needed: Computer Vision / Object Detection (YOLO or similar), OpenCV, RTSP/NVR integration, Python, Dashboard/Web development, some experience with edge or on-prem deployment a plus.

Screening Questions:
- To scope this accurately, walk us through what information you'd need from us before quoting
- at the client needed, what you built, and the outcome).
Have you integrated a CV model with an existing (not new) CCTV/NVR system before?
- Any proof of work sample
- Would you recommend on-prem/edge processing or cloud-based processing for this use case, and why?

Job Title:Interactive "Draw, Scan, Play" Unity System (Computer Vision + Dynamic Sprite Ingestion)Project OverviewWe are looking for a developer or small agency to build a custom interactive educational system for our makerspace. The goal is to create a seamless hardware-software pipeline where a student draws a character or object on a physical sheet of paper, places it under a fixed overhead webcam, and presses a button to scan it. The system must automatically isolate the drawing and dynamically inject their physical artwork as the playable character inside a custom Unity game template.The final system will be a standalone desktop application (running on Windows or macOS) attached to a connected touchscreen monitor and overhead USB camera. It must also generate a QR code so users can take their web-based games home.Technical Scope & WorkflowThe software pipeline must execute the following steps seamlessly:1. Image Capture & Processing (Computer Vision)Input: Capture a high-resolution image from a fixed overhead USB webcam or document camera.Segmentation: Use computer vision (OpenCV or standard C# thresholding scripts) to automatically detect paper boundaries, correct perspective distortion, perform thresholding to completely erase the white paper background, and isolate the drawing.Output: Save the isolated drawing as a transparent .png sprite into a designated directory.2. Game Customization & Asset Mapping (Unity Engine)Templates: Build three (3) simple, pre-coded 2D game templates (e.g., an Infinite Runner, a Platformer, and a Flappy Bird/Jumper clone).Asset Mapping: The selected template must dynamically fetch the newly generated .png sprite from the local directory at runtime and map it as the main playable character with a fitted 2D box/capsule collider.Customization Dashboard Screen: Before the game launches, a touchscreen dashboard menu must allow the user to select modular customization variables that alter the template:Accessories: Choose a digital sticker sprite overlay (e.g., a hat, sunglasses, or cape) that automatically anchors to the top bounds of the scanned character.Audio: Select from 3 pre-loaded background music tracks.Variables: Toggle between basic movement modifiers (e.g., Speed: Normal vs. Fast).3. Deployment & Portability (QR Code Output)Local Play: The game must launch instantly in full-screen mode for immediate local arcade-style gameplay.Mobile Portability: Upon game completion, the system must generate a unique QR code. When scanned by a smartphone, this QR code must open a mobile-optimized WebGL version of the exact game played, hosted on a cloud server (using a free-tier hosting framework like Firebase or GitHub Pages to minimize running costs).Key DeliverablesFully functional local application managing the camera control, image processing pipeline, and touchscreen arcade UI dashboard.Three (3) completed Unity game templates fully integrated into the dynamic asset pipeline.Cloud web-hosting pipeline configuration for serving compiled WebGL builds via QR codes.Clean, well-commented Unity project files and source code with full intellectual property ownership handed over to us upon project completion.Multi-language support architecture (Structured text data files for English and Thai; designed so Spanish text can easily be added by us later).

Looking for an experienced developer to build an AI-based UAV detection and tracking system for a custom ArduPilot quadcopter using Raspberry Pi 5 and AI HAT+ 2. Scope includes real-time drone detection, object tracking, MAVLink integration, safe autonomous follow/stand-off navigation, telemetry logging, GCS interface, failsafes, SITL testing, and complete source-code handover.

Skills: Python, Machine Learning (ML), Robotics, Raspberry Pi, Image Processing, OpenCV, Embedded Systems, Computer Vision, Deep Learning, AI Model Development

we need support to implement iot based auto inspection and jewllery counting if your ready we can do multiple projects

I need comprehensive cleaning for a dataset that combines text, numbers, and a sizable collection of image files. Because the source is truly mixed data, the spreadsheets and the pictures require different but coordinated workflows.

For the tabular side, expect CSV, Excel, and the occasional JSON export. I want every column consistently formatted, missing values handled sensibly, outliers flagged, and–most importantly–all duplicates removed so each record is unique and analysis-ready.

The images come in assorted formats and resolutions. Your focus here is strict formatting deduplication: spot identical or near-identical shots, eliminate them, standardise naming, and ensure each remaining file meets a common resolution and colour profile so they slot smoothly into downstream pipelines.

I’m comfortable if you script the job in Python (pandas, NumPy, OpenCV, Pillow) or use equivalent tools in R or Scala; just keep the process reproducible.

Deliverables
• Cleaned and deduplicated tabular data in original formats
• Curated image set with duplicates removed and uniform specs
• Brief report or notebook that documents every cleaning rule, script, and decision

I’ll validate by running your notebook on a fresh copy of the raw data: no duplicate IDs in the tables and no twin images in the folder will be the acceptance gate.

Skills: Excel, Data Cleansing

We run a platform (Python 3.10, FastAPI, Celery, PostgreSQL 16, Redis, one Docker image) on Azure. We want it moved to AWS, following a design we already have. We are looking for one hands-on senior engineer, not an agency, to own the migration end to end with our in-house developer.

The target design (already decided, you will validate it):

Two EC2 instances (prod and staging) running our existing docker-compose stack behind nginx + certbot
RDS PostgreSQL 16, Single-AZ
S3 for media (replacing Azure Blob), ECR (replacing ACR), Secrets Manager
GitHub Actions building to ECR and deploying via SSM Run Command
Everything in Terraform
Deliberately no EKS, load balancer, NAT gateway or ElastiCache; this is a small, cost-sensitive workload

What you will do:

Confirm the design and sizing, write the migration runbook and rollback plan
Terraform the AWS estate (VPC, EC2, RDS, S3, ECR, Secrets Manager, IAM roles, GitHub OIDC, CloudWatch basics)
Application changes in our Python repo: a storage abstraction so the app writes to S3 instead of Azure Blob (about 20 call sites, boto3 already in the project), compose and CI updates, secrets loading
Copy existing media from Azure Blob to S3 with matching keys and rewrite stored URLs in the database
Stand up staging on AWS, prove alarms, e-mails and viewer links end to end with us
Run the production cutover in an overnight NZ maintenance window (DB dump/restore, DNS switch, smoke tests), then a week of hypercare
Rotate all secrets, decommission the Azure resources, hand over runbooks

Must have:

Several production cloud migrations you personally delivered
Strong Terraform and AWS (EC2, RDS, S3, IAM, SSM, ECR, GitHub Actions OIDC)
Comfortable editing a Python/FastAPI/Celery codebase and writing tests, not just infrastructure
PostgreSQL dump/restore and cutover experience with a real rollback plan
Clear written English; you will be writing the runbooks
Availability for the go-live night in NZ time

Nice to have: Amazon Bedrock, Azure OpenAI, video processing (ffmpeg/OpenCV), Kubernetes (we are moving off it, but you need to read the manifests).

How we will work: read-only GitHub access after an NDA, pull requests reviewed by our developer, a shared Slack channel, milestone payments per phase above. We have a written scope with acceptance criteria and a monthly AWS cost target that we will share with shortlisted candidates.

AI-Based Football Video Analytics – Research Project

I’m looking for an AI/ML + Computer Vision developer/researcher to develop an experimental research prototype + research paper on:

“AI-Based Low-Cost Video Analytics System for Assisting Grassroots Football Talent Identification.”

Project Scope

I already have football video footage. The goal is to build a research-level prototype, not a commercial application.

Workflow:
Football Video → Preprocessing → Player Detection (YOLO) → Player Tracking (ByteTrack/BoT-SORT) → Movement/Position Analysis → Feature Extraction → Player Performance Metrics → Analytical Player Profile

Possible metrics include distance, estimated speed, movement intensity, trajectories, field/zone coverage, and other reliably measurable indicators.

Experimental Requirements

The implementation must produce genuine quantitative results, including where applicable:

-Detection: Precision, Recall, F1, mAP
-Tracking: IDF1, ID switches/MOTA
-Performance: FPS, processing time
-Player-wise performance analysis
-Graphs, tables and visualized/annotated video results
-Model/approach comparison where feasible

No fabricated results.

Research Paper Format

Abstract → Keywords → Introduction → Literature Review & Research Gap → Methodology → Dataset & Experimental Setup → Experiments & Results → Comparative Analysis → Discussion → Limitations → Conclusion → Future Work → References

The paper should include system architecture/workflow, literature comparison table, methodology diagrams, experimental tables/graphs, and relevant visual results.

Technology

Python, OpenCV, YOLO/PyTorch, ByteTrack/BoT-SORT or suitable alternatives. Pretrained models are acceptable; no need to build a model from scratch.

Deliverables
-Working prototype + source code
-Experimental results
-Tables/graphs/visualizations
-Architecture/workflow diagram
-Complete research paper
-Proper academic references

Publication Goal:
Target a legitimate peer-reviewed Scopus-indexed venue, preferably a suitable IEEE/Springer or other relevant journal/conference, subject to the final quality and current indexing status.

Important: This is a research prototype + experimental paper, not a full commercial software system. I already have the football footage.

Please apply only if you have experience in Computer Vision/YOLO, tracking, Python, sports analytics and academic research.

Skills: Python, Statistics, Machine Learning (ML), Big Data Sales, Statistical Analysis, Data Science, Data Visualization, Data Analysis, Computer Vision, YOLO

I want to create a personal-grade robotic solution for use in the home. My primary goal is purely robotic in nature: turning a concept into a working device that can move safely around a domestic environment, carry out simple household interactions, and be controlled through a companion mobile or web interface.

Here’s what I already have in mind:
• Core functionality: autonomous navigation (basic obstacle avoidance), voice or app-based commands, and the ability to perform at least one practical task such as fetching light objects or providing room-to-room telepresence.
• Hardware: I’m open to your recommended microcontroller or single-board computer, but it must be affordable and easy to source globally.
• Software: preference for Python or C++ (ROS, OpenCV, or similar libraries welcome) so that future feature additions are straightforward.
• Form factor: compact enough for apartments, aesthetically friendly, and safe around children or pets.

Deliverables
1. Concept design with component list and estimated costs.
2. Schematics, wiring diagrams, and 3-D model files.
3. Firmware / software package with documented source code.
4. A brief video or simulation demonstrating the robot completing its core task.

Acceptance criteria
• The prototype must navigate a simple home-like test layout without collision.
• Commands issued from the companion app trigger the defined task consistently.
• Full build and setup instructions allow another maker to reproduce the result.

If you have experience with personal or home robotics and can take this from idea to demonstrable prototype, I’d love to see how you would approach it.

Skills: C Programming, Python, Microcontroller, Software Architecture, C++ Programming, Robotics, Embedded Systems, Autonomous Navigation

We're building a pilot AI security monitoring system to replace traditional on-site guarding for sites such as depots, yards, commercial premises, and rural land.

We need a system that can:

Connect to IP/CCTV cameras (existing or freelancer-recommended hardware) and motion/perimeter sensors
Use AI/computer vision to distinguish between people, vehicles, and animals — not just raw motion — to cut down false alarms
Send real-time alerts (app notification, SMS, or WhatsApp/Telegram) when it detects a genuine event, e.g. a person on-site after hours
Provide a simple dashboard or live-view where we can check camera feeds and recent alerts remotely
Be built in a way that can scale from one pilot site to multiple sites later

This is a first pilot build — we want something working and reliable rather than over-engineered. Please propose your recommended tech stack (cameras, AI model/service, hosting, alerting method) as part of your bid, and break your price into phases:

Camera/sensor integration + live feed
AI detection layer (person/vehicle/animal classification)
Alerting + dashboard

Ideal freelancer: experience with computer vision (e.g. OpenCV, YOLO, or similar), IoT/camera integration, and comfortable working with a non-technical client to explain trade-offs (cost vs. accuracy vs. complexity).

We're a small UK-based team moving from traditional in-person security guarding into AI-led monitoring, so real-world reliability matters more than flashy features.

Skills: Website Design, HTML, UI / User Interface, Web Development, Web Design, AI Content Creation

We are an engineering firm looking to build a custom AI-powered CAD automation tool. The goal is to automate the drafting and compliance checking of CCTV floor plans according to strict government Security Systems Department (SSD) criteria.
Instead of manual drafting, we need a tool that can ingest a 2D DWG/DXF or Revit floor plan, use spatial AI to automatically place security cameras for 100% coverage, calculate 3D field-of-view (FOV) while detecting wall blockages, and output a fully compliant, dimensioned CAD file with a generated Bill of Quantities (BOQ).
Core Deliverables
Data Extraction: Programmatically parse DWG/DXF/RVT files to identify walls, doors, and room types.
AI Spatial Routing & Camera Placement: Build the logic/algorithm to automatically place cameras at optimal locations (entrances, hallways) to achieve complete coverage without blind spots.
Rules Engine for Compliance: Encode strict parameters (e.g., minimum Pixel-Per-Meter density, privacy zone avoidance, required storage calculations) to ensure the generated design passes a digital compliance check.
Automated Output: Push the optimized data back into the CAD software to automatically draw standard symbols, field-of-view cones, wiring layers, and annotations.
Required Tech Stack
CAD/BIM Integration: Strong experience with Autodesk Platform Services (Forge), AutoCAD API (AutoLISP/.NET), or Revit API (C# or Python).
AI / Spatial Computing: Python, OpenCV (or similar computer vision libraries), and algorithmic geometry for analyzing 2D/3D spaces.
Previous Experience: You must have previously built plugins, scripts, or cloud apps that automate drafting or geometry generation inside CAD environments.

Skills: Python, CAD/CAM, Solidworks, AutoCAD, Lisp, AI Development, CAD / SolidWorks, AI Automation

Research Paper + AI Football Video Analytics Prototype

I am looking for someone to help me develop an experimental research project and research paper titled:

“AI-Based Low-Cost Video Analytics System for Assisting Grassroots Football Talent Identification.”

The goal is to create a research prototype, not a commercial application or a complete production-level software system.

What I need

The system should take a normal football match/training video recorded using a low-cost/ordinary camera as input and use existing/pre-trained AI/computer-vision models to analyze players.

Basic workflow:

Football Video → Player Detection → Player Tracking → Movement/Position Analysis → Performance Metrics → Player Performance Report

The implementation may use existing models such as YOLO, ByteTrack/BoT-SORT, OpenCV, etc. There is no requirement to develop a new AI model from scratch.

Expected analysis

Depending on what can be reliably extracted from the video:

Player detection
Player tracking
Player movement/trajectory
Estimated distance/speed
Movement intensity
Position/field coverage
Other relevant football performance indicators
Player-wise analytical report/score
Research/Experimental Requirement

This is specifically for an experimental research paper, so I need actual results from the implementation, including appropriate evaluation metrics such as detection accuracy, precision, recall, F1/mAP, tracking metrics, FPS/processing time, etc.

The final paper should include:

Literature review & research gap
Methodology
System architecture
Dataset/video description
Experimental setup
Experiments
Results with genuine data
Tables and graphs
Discussion
Limitations
Conclusion & future work
Proper academic references

Important: I do NOT need a large commercial software product, mobile app, website, login system, or cloud deployment. The priority is a working research prototype + genuine experimental results + publication-quality research paper.

Please mention your experience with computer vision, YOLO/object detection, multi-object tracking, sports analytics, and research-paper writing when applying.

Skills: C Programming, Python, Software Architecture, Research Writing, Machine Learning (ML), C++ Programming, Computer Vision, YOLO

I have a fully-working Android Studio project (Java) that photographs urine test strips and reads the colours, but its interpretation layer is letting us down: colours are detected correctly, yet the final numbers it returns are wrong. I need someone to dig into the existing code, locate the flaw in the result-interpretation logic, and at the same time refine the overall algorithm so the readings become clinically reliable.

Because I’m interested in both fixing the bug and improving the core method, you will be free to refactor, recalibrate, or even redesign parts of the analysis pipeline provided the camera capture and colour-extraction steps remain intact. OpenCV and native Android APIs are already integrated, so you will be working inside that framework.

Key deliverables:
• Diagnose and eliminate the logic error that produces incorrect results.
• Rework the interpretation algorithm so that, when tested against my reference images, each parameter reports within the manufacturer’s stated tolerance.
• Supply concise comments plus a brief changelog explaining the adjustments.
• Push a clean, buildable Android Studio project back to my private Git repository.

Skills: Java, Mobile App Development, Android, Algorithm, Software Architecture, Software Development, Image Processing, OpenCV, Android App Development, Android Studio

I need a robust automation bot for an Android game that can do three core things:

1. Detect every relevant UI element on multiple screens through image recognition, then tap the right spots to claim the daily rewards without fail.
2. Autonomously solve the in-game maze and pick-pair minigames with human-like speed and accuracy.
3. Run hands-free once launched, logging what it collected or solved so I can review a simple report afterward.

Deliverables
• Full Android-ready build (APK or script + instructions to run with ADB/Emulator)
• Source code with clear comments
• Image libraries or model files used for recognition
• Quick user guide and a short demo video that shows the bot clearing both minigames and collecting a day’s rewards

Acceptance criteria
• ≥100% success rate in detecting reward buttons across all tested devices (1080p & 720p)
• <5 seconds average solve time for each maze or pick-pair round
• No crashes after hours continuous looping

Tools you pick are up to you—OpenCV, TensorFlow Lite, AutoInput, or a custom accessibility service are all fine as long as the end result meets the above metrics.

Clear, fluent English is essential. When I’m online and i ask some when u get online and read it u will respond no leave me on read I expect prompt chat replies, and if something is complex I may request a quick voice call through Freelancer to sort it out faster.

Skills: Java, Mobile App Development, Android, Software Architecture, OpenCV, Automation, Android App Development, Image Recognition

We are seeking an experienced Embedded Systems and Edge AI Engineer to develop a functional proof-of-concept for an indoor visual and thermal monitoring device.

This is an early-stage prototype—not a mass-production design or certified fire/life-safety product. The objective is to demonstrate reliable sensor integration, local AI processing, configurable detection zones, thermal anomaly detection, and intelligent alerts.

CORE MVP FUNCTIONS

1. Zone intrusion detection
Detect a person entering a user-defined region of interest using a standard RGB camera and an edge-based computer vision model.
2. Thermal anomaly detection
Use a radiometric-capable thermal sensor or camera to identify:

* A temperature exceeding a configurable threshold.
* An abnormal rate of temperature increase.
* A hot spot deviating from its established baseline.

3. Intelligent alerts
Generate an alert containing:

* RGB image.
* Thermal image or heat map.
* Timestamp.
* Measured temperature.
* Detection type.
* Confidence score.

TECHNICAL DIRECTION

The prototype should use commercially available components and local edge processing. A custom PCB is not required for this phase.

Preferred architecture:

* NVIDIA Jetson or Raspberry Pi-class edge computer.
* RGB camera.
* Radiometric-capable thermal sensor/camera.
* Ethernet/PoE preferred.
* Wi-Fi may be included for testing.
* Local storage.
* Basic web dashboard for viewing events, configuring detection zones and adjusting thermal thresholds.
* Compact 3D-printed enclosure suitable for installation near an indoor wall/ceiling corner.

The engineer may recommend alternative components if they provide better performance, availability or cost efficiency. No hardware may be purchased without written client approval of the bill of materials and pricing.

EXPECTED DELIVERABLES

* System architecture and component selection.
* Detailed bill of materials with suppliers and pricing.
* Functional bench prototype integrating RGB and thermal sensors.
* Person detection within a configurable zone.
* Thermal threshold, rate-of-rise and baseline-deviation detection.
* Alert generation with visual and thermal evidence.
* Basic local web dashboard.
* Editable enclosure CAD files and a 3D-printed prototype enclosure.
* Fully assembled and functional physical prototype shipped to Florida, USA.
* Complete source code, configuration files and build instructions.
* Setup, testing and troubleshooting documentation.
* Final recorded demonstration and live video handoff session.

PROJECT REQUIREMENTS

* Demonstrated embedded systems and hardware integration experience.
* Experience with computer vision and edge AI.
* Experience integrating thermal sensors or thermal cameras.
* Python, OpenCV and Linux proficiency.
* Experience with Jetson, Raspberry Pi or comparable platforms.
* Ability to design or coordinate a basic 3D-printable enclosure.
* Clear English communication.
* Weekly progress demonstrations.
* Ability to ship the completed prototype to Florida, USA.

PROJECT STRUCTURE

This will be a fixed-price project completed through milestones. The current engineering budget is approximately $3,500 USD. Prototype components and approved shipping expenses will be handled separately, subject to prior written approval. The expected timeline is 8–12 weeks.

All source code, CAD files, documentation, configurations and other work product created specifically for this project must be delivered to and owned by the client upon payment. Additional confidential product details will be shared only with the selected candidate.

This prototype is an experimental monitoring and early-warning system. It is not intended to replace certified smoke detectors, fire alarms or other regulated life-safety equipment.

HOW TO APPLY

Please begin your proposal with the words “THERMAL EDGE” and answer the following:

1. Describe a relevant project where you integrated cameras, sensors and an edge computing platform.
2. Which thermal sensor/camera and edge computer would you recommend for this prototype, and why?
3. How would you detect both absolute temperature thresholds and abnormal temperature rise over time?
4. Can you deliver the physical prototype, source code, editable CAD files and documentation within 8–12 weeks?
5. State your current country and explain how you would ship the finished prototype to Florida.
6. Provide an initial milestone breakdown for the $3,500 engineering budget.

Generic proposals that do not address these questions will not be considered.

We are an established saas software company serving the pavement industry and are looking to develop an AI-powered aerial takeoff tool.

The goal is simple:

A user enters a commercial property address, the system retrieves aerial imagery, automatically identifies pavement-related areas and objects, calculates measurements, and allows the user to edit or confirm the results.

### Initial Phase

For Phase 1, we want to focus primarily on:

* Asphalt pavement square footage
* Concrete square footage
* Parking stalls
* ADA stalls
* Curbs
* Islands
* Sidewalks
* Striping

The system should display the property on an aerial map, create editable polygons or measurements, and return accurate real-world quantities.

Eventually, these measurements will integrate directly into our existing pavement estimating software.

### What We Are Looking For

We are specifically looking for someone with strong experience in:

* Computer vision
* Aerial or satellite imagery
* Image segmentation
* Object detection
* GIS / geospatial mapping
* Python
* PyTorch or similar AI frameworks
* OpenCV
* GeoJSON / PostGIS
* Mapbox, Nearmap, or similar imagery APIs

Experience building measurement, mapping, construction takeoff, property intelligence, roofing, paving, landscaping, or similar geospatial applications would be especially valuable.

### First Milestone

Our first goal is not to build the entire platform.

We want to prove that the technology works.

The first milestone would be:

**Enter a commercial property address → retrieve aerial imagery → automatically identify the asphalt area → calculate square footage → display an editable boundary around the detected pavement.**

If this works successfully, we would move into additional pavement features and integration with our existing software.

### When Applying

Please send:

1. Examples of similar computer vision, aerial imagery, GIS, or mapping projects you have completed.
2. A brief description of how you would approach this project.
3. The technologies you would recommend.
4. Your estimated cost for an initial proof of concept.
5. Whether you personally will be doing the development or if this will be handled by a team.

We are looking for someone who can help us determine the best technical approach, not simply someone who can write basic application code.

This could become a significant long-term project for the right person or team.

I’m building a computer-vision pipeline focused on reliable detection and recognition of human faces and full-body figures. The core of the job is a clean, well-documented Python implementation that can take still images or short video clips and return bounding boxes, class labels, and confidence scores for each detected person.

I already have test media and the computing environment; what’s missing is the detection logic itself—ideally leveraging familiar libraries such as OpenCV, TensorFlow, PyTorch, or a proven YOLO/SSD variant. Accuracy on varied lighting and crowded scenes is more important to me than sheer speed, but the code should still run in real time on a modern GPU.

Deliverables
• Python source code with clear inline comments
• Pre-trained weights (or training notebook) and instructions for further fine-tuning
• Short report outlining model choice, evaluation metrics, and sample results
• Simple CLI or notebook demo that shows the system working on my test set

Acceptance criteria
• ≥90 % precision and recall on the provided images
• No external dependencies beyond standard Python CV/ML stacks
• Reproducible setup: a requirements.txt or environment.yml that builds without errors

Once this module is solid, there may be follow-on tasks for simulation apps and thesis-level documentation, but for now the priority is a robust, plug-and-play face-and-body detector.you are cute

Skills: Python, Software Architecture, CUDA, Machine Learning (ML), Image Processing, OpenCV, Computer Vision, Deep Learning, Object Detection, YOLO

Hi, I’m looking for an experienced developer or small team to build an AI-powered video editing platform that combines AI-assisted editing with manual video editing. I’m looking for strong experience in AI integration, video processing, FFmpeg, backend development, and scalable cloud infrastructure. Full product details and workflow will be shared only with shortlisted candidates after an NDA is agreed upon. Please provide examples of relevant AI and video projects you have built.

Tengo un sistema de automatización ya operativo y necesito a alguien que lo lleve al siguiente nivel.
La base está construida en Python y OpenCV, comunicándose con un ESP32; todo corre sin errores, pero falta completar la lógica de detección de objetos y la contabilización de productos.

Lo que requiero:
• Ajustar y optimizar el modelo de detección de productos (actualmente uso OpenCV + Python).
• Implementar el conteo automático de cada producto que aparezca en la cámara.
• Enviar el resultado al ESP32 a través de la interfaz existente para que el microcontrolador lo procese.
• Dejar el código limpio, documentado y con instrucciones rápidas para volver a entrenar o cambiar el modelo si se añaden nuevos productos.
• Probar la solución en vídeo en tiempo real y entregar un breve informe con métricas de precisión y velocidad.

Dispongo de acceso remoto al hardware y puedo proporcionar secuencias de vídeo de prueba. Si dominas Python, computer vision y la comunicación serie/Wi-Fi con ESP32, me encantará conocer tu propuesta y plazos.

Skills: Python, Software Architecture, Machine Learning (ML), C++ Programming, OpenCV, Computer Vision, Automation, Object Detection

I’m currently working on an AI Multicam Editor plugin for Adobe Premiere Pro. The plugin reads the active sequence and the cameras placed on tracks such as V1, V2, V3, and V4, then sends the video and timeline data to a local engine running on the computer. The engine analyzes the footage using AI and tools such as YOLO, OpenCV, FFmpeg, and PySceneDetect, evaluating each camera based on factors like people detection, sharpness, shake, framing, and scene changes. It then chooses the best camera for each moment, builds an automatic editing plan, and applies the decisions back inside Premiere. It also includes features such as a Sequence Queue and a Smart Transition Director.

At the moment, I’m having some issues with the decision-making part of the system, especially when it comes to choosing the right shot, the right timing, and the right duration for each cut. Sometimes the system selects an angle or shot length that does not fit the scene well enough. I also need it to become smarter at ignoring unwanted mistakes and bad movements, such as camera shake, random motion, weak shots, awkward transitions, or anything that reduces the quality of the edit.

My goal is for the system to understand the scene better, choose the most appropriate camera at the most appropriate moment, use natural shot durations, avoid bad or distracting footage, and produce an editing rhythm that feels as close as possible to the work of a professional video editor, especially in terms of shot selection, cut timing, continuity, camera changes, and the overall flow of the final edit.

I need a complete mobile app that runs smoothly on both iOS and Android and showcases advanced facial-recognition capabilities. The core of the build is a camera workflow that can:

• Face identification
• Emotion detection
• Age and gender estimation

Those three features must fire quickly on-device (or via a lightweight cloud microservice if latency is acceptable) and return an easy-to-parse JSON result I can feed into future modules. The interface should feel modern and sleek with clean typography, subtle animations and dark-mode support.

You’ll own the full stack—from selecting the best SDK (e.g., Apple Vision, Google ML Kit, OpenCV, or a custom model in TensorFlow Lite / Core ML) to wiring up the UI in Swift/Kotlin or a unified framework like Flutter or React Native, whichever yields the best performance. Source code, build scripts, documented APIs, and a testable release build for each platform are expected at hand-off.

I have budgeted roughly £15 per hour; we can break that into clear milestones once you map out the sprint schedule. Let me know your preferred toolchain and any similar work you’ve shipped so we can get started right away.

Skills: Mobile App Development, iPhone, Android, Objective C, OpenCV, React Native, Flutter, Facial Recognition

Job Description
Overview
We are a regional property preservation company working with clients such as MCS, Freddie Mac, Guardian, Altisource, NFR, and direct bank relationships. Our field operations run on 1099 independent contractors who complete work orders (initial secures, winterizations, grass cuts, bid approvals/requests, and dozens of other property preservation tasks) and upload photo/measurement documentation into our industry-standard system, Property Preservation Wizard (PPW).

Today, a team of in-office client managers and overseas processors manually reviews contractor photos, builds bid sheets, submits bids to clients, reviews after-photos once work is approved and completed, and marks orders "ready for office" for invoicing. This work is high-volume, repetitive, and photo/data-driven — and we believe it's a strong candidate for AI-based automation.

What We're Looking For

We want to build (or integrate) an AI-powered system that can take over the manual, judgment-based review work currently done by our processing team, including:

1. Photo Review & Compliance Check — Automatically review contractor-submitted before-photos against work order instructions/notes to confirm scope, condition, and completeness before bidding.
2. Automated Bid Generation — Analyze photos, measurements, and work order type to generate an accurate bid sheet, using historical pricing data and geographic/market pricing benchmarks to recommend competitive, defensible pricing.
3. Bid Review/QA — A secondary check layer that flags bids that look inconsistent, underpriced, overpriced, or incomplete before submission to the client.
4. Post-Completion Photo Verification — After contractors complete approved work and upload after-photos, automatically compare against the approved bid scope to confirm the work matches what was bid and approved.
5. Exception Flagging — Identify likely follow-ups, missing items, or client rejection risks proactively, so our team can address them before client pushback rather than after.
6. "Ready for Office" Automation — Where verification passes, automatically route the work order to ready-for-office/invoicing status without manual sign-off.
7. Automated Bid Approval Pricing — Review approved bids and automatically assign contractor pricing based on client requirements, historical costs, market conditions, and target margin thresholds.

Ideal Technical Approach

We are open to your recommended architecture, but anticipate this involves:
• Computer vision / image analysis (to assess property condition and verify completed work from photos)
• LLM-based reasoning over work order notes, client instructions, and historical data to generate bid line items and pricing
• Integration with our PPW portal (API integration, RPA/browser automation, or a middleware layer — TBD based on what PPW exposes)
• A pricing/market-rate database or model that can be trained or fine-tuned on our historical bid data by geography and work order type
• A workflow/orchestration layer to move work orders through review → bid → submission → post-completion QC without manual intervention (with human-in-the-loop override/approval where needed)

Scope of Engagement

We'd like to start with a discovery/scoping phase to assess:
• What PPW allows in terms of API access, data export, or automation hooks
• What data we have available (historical bids, photos, pricing outcomes) to train or ground the system
• A recommended architecture and phased build plan (e.g., starting with photo review or bid generation before full end-to-end automation)
From there, we're looking for a developer/team to build a working system in phases, ideally starting with a pilot on a limited set of work order types before expanding.

What You'll Need
• Strong experience with AI/LLM integration (e.g., Claude, GPT, or similar) and computer vision/image analysis
• Experience building workflow automation or RPA systems that integrate with third-party portals/systems
• Ability to work with messy, real-world business data (historical bids, photos, notes) and structure it for model use
• Comfort scoping an ambiguous, real-world operations problem into a phased technical plan — we are not handing you a finished spec, we're handing you a business process
• Experience with (or willingness to quickly learn) property preservation, field services, insurance/claims, or similar photo-and-bid-driven industries is a strong plus but not required

How to Apply

Please share:
• Relevant experience with AI/computer vision automation projects, especially any involving photo verification, bid/pricing automation, or workflow integration with third-party systems
• Your recommended high-level approach to a discovery phase for a project like this
• Availability and whether you work independently or as part of a team

I run several AI initiatives that revolve around computer-vision pipelines, and I need an adaptable hand to keep our image data in perfect shape before it ever touches a model. Your main focus will be classic machine-learning groundwork: pulling raw image assets from cloud storage, auditing their quality, handling augmentations, labeling inconsistencies, and packaging everything into tidy, well-documented datasets that flow straight into our training scripts.

Most tasks live in the data-preprocessing and cleaning stage, so you should be comfortable writing reproducible code for resizing, normalization, class-balancing, and automated sanity checks. We currently use Python with Pandas, NumPy, OpenCV, and sometimes Albumentations; if you have a favorite toolkit that speeds things up, I am open to it as long as it’s clearly explained and container-ready.

Once a batch is processed, you’ll push it to our Git-based repo and drop a short markdown report that lists what changed, any edge cases you spotted, and the commands needed to reproduce the run. Clean commits, readable notebooks or scripts, and strict version control are non-negotiable; they keep the downstream modeling and monitoring steps painless.

If this mix of autonomous, image-centric ML work suits you, let me know your typical turnaround time for a ten-thousand-image dataset and highlight a past project where your preprocessing directly improved model performance.

Skills: Java, JavaScript, Python, Data Processing, Software Architecture, Machine Learning (ML), OpenCV, Data Annotation

Overview

Looking for a Computer Vision Engineer to build a Python-based pipeline for Optical Braille Recognition (OBR) — processing binary/thresholded images of Braille dot arrays (sample attached) and converting them into structured text.

Key Responsibilities

- Preprocessing & Grid Alignment: Detect, segment, and align the grid structure of Braille dots from binary/preprocessed images.
- Dot Pattern Extraction: Group detected dots into standard 2×3 or 2×4 Braille cell structures, accounting for skew, spacing variability, and noise.
- Decoding Pipeline: Map extracted dot patterns to Unicode/ASCII text using standard Braille translation tables.
- Evaluation: Deliver a structured, reproducible notebook/script with accuracy metrics, detection performance, and processing speed.

Required Qualifications

- Strong Python, OpenCV, and image processing (scikit-image, NumPy, SciPy).
- Experience with classical CV techniques (contour analysis, Hough transforms, connected component analysis, morphological ops) or lightweight deep learning detection (YOLO, custom CNNs).
- Familiarity with OCR or document analysis pipelines.

Deliverables

- Working code with setup/run instructions.
- Notebook or report with accuracy metrics and performance evaluation.