Find OpenCV Projects on Upwork with Vollna

Senior Embedded Vision / Data Capture Engineer (NOT a model training role)

🔍 Summary

We are building a road inspection device that records high-quality video + GNSS/IMU data for offline AI analysis.

We are not looking for a PyTorch/TensorFlow researcher.
We need an engineer who can design and implement a robust embedded video + sensor logging pipeline on an edge device.

If your primary experience is training YOLO models in Python, this role is not for you.


---

🎯 What this role is really about

Your main mission:

Take an industrial camera + GNSS + IMU + embedded board,

And turn it into a reliable “black box” recorder:

1080p video at stable 25–30 FPS,

synchronized GNSS/IMU timestamps,

data safely logged to SSD,

no frame drops, no random exposure jumps.



AI will run offline later.
Your focus is data capture, synchronization, and robustness, not model training.


---

✅ Required Core Skills (must-have, no exceptions)

Embedded Linux & C++

Strong C++ on Linux (not just Python/OpenCV scripts).

GStreamer and/or V4L2 for building low-latency camera pipelines.

Experience with:

multi-threading,

ring buffers,

back-pressure handling,

handling dropped frames, etc.



Industrial Camera Integration

Hands-on experience with USB3 / MIPI / GMSL industrial cameras.

Ability to configure:

exposure,

gain,

shutter,

resolution,

frame rate programmatically.


Understanding of rolling vs global shutter and impact on motion.


Sensor & Timestamp Handling

Reading GNSS (NMEA/UBX or similar) and IMU data streams.

Attaching accurate timestamps to each sensor sample.

Aligning sensor data with video frames (same clock / known latency).


Edge Device Engineering

Experience with Jetson Orin / Xavier / similar ARM-based embedded devices.

Designing video logging systems that:

write to SSD,

survive reboots / power glitches,

recover gracefully from errors.


Comfortable with Linux services, logging, debugging tools.


System-level Thinking

You have built real systems in the field, not just lab demos.

Can define clean interfaces for a hardware manufacturer:

“This is the API/connector/protocol we need”.


Can explain trade-offs: bitrate vs quality vs storage vs CPU load.



---

👍 Nice-to-have (NOT required, but helpful)

Experience with:

automotive or mobile robotics perception,

road / infrastructure imaging,

Shenzhen-based camera or box-build manufacturers.


Basic familiarity with TensorRT/ONNXRuntime only to the extent needed to run a pre-trained model (we can keep this minimal for MVP).



---

❌ This role is NOT for you if…

Please do NOT apply if:

Your main skill is:

PyTorch/TensorFlow model training,

computer vision research,

Kaggle-style experimentation.


You mostly work in:

Jupyter notebooks,

Google Colab,

cloud-only pipelines.


You have never:

built a real-time camera pipeline,

debugged dropped frames or sync issues,

worked directly with industrial cameras.



We respect ML/AI researchers — but this is an embedded systems & data pipeline role.


---

📦 Key Deliverables (January MVP)

For the first MVP (by end of January), we expect:

1. Working prototype on dev kit

1080p video capture at stable 25–30 FPS.

Known and measured latency from sensor to disk.



2. Synchronized logging

Video + GNSS (+ IMU if available) written with usable timestamps.

Simple format (e.g. MP4 + JSON/CSV metadata or similar).



3. Basic configuration tooling

Ability to adjust key camera params (exposure, gain, FPS).

Simple CLI or minimal UI for debugging & verification.



4. Documentation

Clear description of pipeline architecture.

Requirements for the hardware manufacturer (camera/box builder).




You will work closely with a Chinese camera manufacturer (hardware/firmware) and the product owner. Good written communication is important.

I need an automation tool in Python that works on Windows and connects directly to DXCAM (DirectX-based screen capture) to monitor a 'Hold & Win' slot machine in real time. (The game will be run in the browser).

The program must:
• Capture the game window at full speed through DXCAM, without noticeable impact on gameplay.
• Detect the MINI, MINOR, MAJOR, and GRAND symbols the moment they appear and track them as they move across the reels.
• Calculate when the token is at the center of the reel and send a precise automatic 'Stop' command.
• The program must be able to stop the MINI, MINOR, MAJOR, and GRAND tokens.

Key points to consider for the program:
• Performance is critical: the symbols fall quickly, so latency from capture to the stop action must be minimized. • Multithreading to avoid latency and freezes.
• Use any support library you want (for example, OpenCV, NumPy, Torch) as long as the setup remains a single command through requirements.txt.

Deliverables
1. Fully documented Python source code.
2. A README with setup steps, configurable parameters (for example, grid size, hotkey assignments), and troubleshooting tips.
3. Demo video and live session showing that the tool stops in time.

Acceptance criteria will be a live test session in the target game; the script must achieve at least 90% effectiveness.

Skills: Python, Image Processing, OpenCV, Computer Vision, Automation, Machine Learning Algorithms

I need an automation tool in Python that works on Windows and connects directly to DXCAM (DirectX-based screen capture) to monitor a 'Hold & Win' slot machine in real time. (The game will be run in the browser).

The program must:
• Capture the game window at full speed through DXCAM, without noticeable impact on gameplay.
• Detect the MINI, MINOR, MAJOR, and GRAND symbols the moment they appear and track them as they move across the reels.
• Calculate when the token is at the center of the reel and send a precise automatic 'Stop' command.
• The program must be able to stop the MINI, MINOR, MAJOR, and GRAND tokens.

Key points to consider for the program:
• Performance is critical: the symbols fall quickly, so latency from capture to the stop action must be minimized. • Multithreading to avoid latency and freezes.
• Use any support library you want (for example, OpenCV, NumPy, Torch) as long as the setup remains a single command through requirements.txt.

Deliverables
1. Fully documented Python source code.
2. A README with setup steps, configurable parameters (for example, grid size, hotkey assignments), and troubleshooting tips.
3. Short demo video or GIF showing that the tool stops in time.

Acceptance criteria will be a live test session in the target game; the script must achieve at least 90% effectiveness.

Skills: C Programming, Python, Software Architecture, C++ Programming, Image Processing, OpenCV, Computer Vision, Automation

We’re looking for an expert in video streaming and surveillance systems to architect and build high-performance, low-latency video pipelines. You’ll work with C++, FFmpeg, and GStreamer to integrate IP cameras, optimize live ingest, and deliver reliable OTT and CCTV workflows at scale. This role includes designing NVR/DVR systems, implementing AI-powered analytics, managing AWS MediaLive/MediaTailor pipelines, and ensuring secure delivery with DRM.

You’ll lead technical decisions, improve infrastructure, and collaborate with product, AI/ML, and cloud teams to ship video systems that actually perform in the real world. If you thrive building robust streaming and surveillance solutions from ingest to delivery, we’d love to work with you..

YOLO Expert: Crowd Counting from Handheld Video (Stabilization Required)

Project Overview
I am a Data Artist looking for a technical partner to extract crowd data from a video recording of a protest march. I have a handheld video taken from a single location of a continuous stream of people walking past a fixed point.

My goal is to objectively measure the number of attendees, as information on recent protester numbers is hard to attain. I need to convert this visual evidence into a rigorous dataset that I will later sonify (turn into sound).

The Assets
* Input Media: 3 minutes long, 1080p, MP4 file.
* Conditions: Handheld shot from sidewalk.
* CRITICAL NOTE: This footage was NOT shot on a tripod. It contains natural camera drift. Your solution must include stabilization/registration logic (e.g. Optical Flow) to ensure the counting line remains fixed relative to the street.

The Task
Use an object detection model (e.g., YOLO v8/v10 or similar) to analyze my video file and provide the following:

1. Total Unique Headcount: An estimated total count of individuals who passed the camera during the duration of the clip. (Note: Since this is a continuous stream of people walking past, the solution likely requires "line crossing" logic or object tracking/Re-ID to avoid double-counting the same people across frames).

2. Time-Series Data (Crucial): I need the data export to be granular so I can map it to sound parameters.

Required Deliverables
1. The Dataset (.CSV): A spreadsheet with high-resolution temporal data. I specifically need these three columns to separate Crowd Size from Crowd Speed:
* Timestamp (or Frame Number)
* People_In_Frame (Crucial: The count of people visible on screen at that moment. This measures the Density).
* Flow_Rate (People crossing the line per second. This measures the Speed/Movement).
* Cumulative_Count (Total unique people counted).

2. Visual Verification (.MP4): A rendered clip (at least 30 seconds) showing the bounding boxes and the stabilized counting line. This is essential for me to verify the rigour of the count against the camera motion.

Questions for You
1. Stabilization Strategy: Given the footage is handheld, how will you ensure the counting line stays registered to the background/street to prevent counting errors? (e.g., Optical Flow, feature matching, or stabilization pre-processing?)
2. Occlusion: How will your approach handle people blocking each other in a dense crowd?
3. Data Rate: Can you output the CSV data at a frequency of 10 readings per second (or per frame)? I need high-resolution control data for granular audio synthesis.

Budget & Proposal Structure
I am looking for a quick turnaround (48-72 hours).

Quote: Please provide a fixed-price quote for the full deliverable (Stabilization + Count + CSV Export).

Itemization: In your technical workflow, the "Total Count" usually requires tracking individuals over time anyway. However, if generating the detailed granular CSV adds significant manual effort or compute cost to your specific workflow, please specify this in your proposal so we can discuss.

• Have you worked with Optical Flow or camera stabilization for object counting before?
• Please paste a link to a similar project or video verification you have created.
• Can you deliver the specific CSV columns (Flow Rate and Density) requested?

Description:

We are a Canadian travel company that receives scanned and handwritten PDF surveys from travellers after their tours. Each survey has a consistent 7 page layout and contains traveller names, tour metadata, multiple rating grids with checkboxes 1 to 5, several open text sections, and a final page with two handwritten questions about destinations and future travel plans.

We need an experienced AI, OCR, or Computer Vision developer to build a reliable and accurate system that extracts all survey data into structured JSON or CSV. This requires someone familiar with layout detection, checkbox mark detection, image processing, and handling noisy or inconsistent scans.

Project Requirements:

Input
A PDF that may contain one or more survey responses. Each survey consists of 7 pages with the same structure.

Fields to Extract
Identity and metadata such as traveller names, tour code, guide names, and start or end dates.
Multiple rating grids where each row is scored from 1 to 5. You must detect which checkbox is marked even if the mark is messy, faint, or not recognized as text.
Open text answers spread across several pages.
Final page contains two text questions about other destinations and confirmed travel plans for future years such as 2025 to 2027.

Output
A structured JSON object or CSV row containing all extracted fields, including ratings, text answers, and identity fields.

Technical Approach
You may use any tools that ensure accuracy, such as Google Document AI, Google Vision plus OpenCV, Tesseract plus segmentation, LayoutParser, or a custom approach. The key goal is reliability.

Required Capabilities
Accurate checkbox or tick detection for 1 to 5 rating scales.
Consistent layout mapping so each extracted value maps to the correct field.
Separation of multiple surveys in a single PDF.
Clear extraction of open text fields.
Error handling and optional debug output such as bounding box or segmentation logs.
Clean code and documentation.

Success Criteria
All rating fields produce correct numeric values.
All open text fields are extracted reliably.
The system works across multiple scans of the same form template.
The tool can process more than one survey per PDF.
Low error rate and stable output.

Ideal Candidate
Experience building OCR or computer vision systems.
Comfortable with form recognition, checkbox detection, and image processing.
Experience with OpenCV, Tesseract, DocAI, or similar tools.
Understands bounding boxes, segmentation, and dealing with low quality scans.
Able to design a reliable extraction workflow suitable for production use.

Deliverables
A working survey processing tool or module.
Source code and documentation.
Instructions for adjusting or adding field mappings.
Optional assistance deploying into a Google Cloud Function.

I’m working with a collection of images that come paired with captions or other textual metadata, and I need a concise, well-structured Python solution that lets me analyse both components in one pass. The image side will involve standard loading and preprocessing; the text side must run sentiment analysis, text classification, and targeted data extraction on the accompanying words.

Feel free to tap into familiar toolkits such as OpenCV or Pillow for the visuals, and spaCy, NLTK, or Hugging Face transformers for the NLP work—whatever you are most productive with, as long as the dependencies are clearly listed in a requirements.txt.

Deliverables
• A modular Python script (or Jupyter notebook) that ingests a folder of images plus their text, cleans and prepares each modality, and produces:
 1. Structured image features or summaries that I can reuse later.
 2. Sentiment scores, classification labels, and extracted key data points for every text item.
 3. A consolidated CSV or JSON report combining the above outputs.
• Clear inline docstrings, a brief README that explains how to run everything from the command line, and a small sample dataset to prove it works.

Acceptance criteria
• I can install via pip -r requirements.txt without errors.
• Running one command processes the supplied sample and generates the final report files in an /output directory.
• Results are reproducible on Python 3.9+.

Keep the code readable and commentary light—once the prototype works I’ll extend it further myself.

Skills: Python, Software Architecture, Machine Learning (ML), Image Processing, Data Extraction, Data Analysis, Natural Language Processing, Sentiment Analysis

Looking for a full stack developer/manager to handle the development of a local AI application that will be used for metal fabrication.

The software (at minimum) will be able to:
-Upload PDF RFP → extracts quantity, material, sizes, tolerances
-Auto-nests parts on sheet (95%+ material yield)
-Real-time design checks (bend radius, hole issues, etc.)
-One-click G-code export for plasma/laser/waterjet
-Installer package for fresh PCs

Tech Stack:
Python 3.13, Electron + React (or PyQt/Tauri), FastAPI backend, spaCy, OpenCV, PyTorch (local inference), DEAP/RL nesting, PyInstaller

MUST HAVE:
5+ years Python + desktop app experience
Shipped 2+ production AI desktop tools (links required)
Strong with local LLMs, OpenCV, nesting/geometry algos
Available 20+ hrs/week, 2-3x a week Zoom updates
Previous CAD, solid works and/or metal fabrication drafting experience

• MOST IMPORTANT: Your experience with metal fabrication software.
• Describe your recent experience with similar projects
• Please list any certifications related to this project
• What frameworks have you worked with?

I need a desktop application that runs natively on Linux and lets users open, access, and manage a “money-growing” wallet without ever presenting identity documents. The only gatekeeper should be biometric authentication: Face ID or Touch ID, depending on the hardware the user has available.

At launch the flow must feel effortless—open the app, present a face or fingerprint, and land on a dashboard showing current balance, growth rate, and a simple button to add or withdraw funds. Under the hood you are free to pick the most reliable open-source libraries (for example OpenCV or libfprint) as long as the recognition step is fast and the data never leaves the user’s machine unencrypted.

Because I’m working on a tight timeline, I’m hoping to see tangible progress right away and a first runnable build as soon as you can ship one. You’ll impress me most if your proposal highlights previous Linux desktop apps you have delivered, especially anything involving biometrics or crypto/fin-tech security.

Deliverables (acceptance criteria)
• Installable .deb or AppImage plus complete source code
• Biometric onboarding & login that functions offline
• Encrypted local wallet capable of deposit, withdrawal, and an interest-style growth simulation
• README with build steps and a quick “getting started” guide

I’m ready to move forward immediately, so if you have solid experience in Linux desktop development and biometric integration, let’s talk right away.

Skills: Linux, User Interface / IA, Cryptography, Software Development, OpenCV, App Development, Software Engineering, Desktop Application

I need a desktop application that runs natively on Linux and lets users open, access, and manage a “money-growing” wallet without ever presenting identity documents. The only gatekeeper should be biometric authentication: Face ID or Touch ID, depending on the hardware the user has available.

At launch the flow must feel effortless—open the app, present a face or fingerprint, and land on a dashboard showing current balance, growth rate, and a simple button to add or withdraw funds. Under the hood you are free to pick the most reliable open-source libraries (for example OpenCV or libfprint) as long as the recognition step is fast and the data never leaves the user’s machine unencrypted.

Because I’m working on a tight timeline, I’m hoping to see tangible progress right away and a first runnable build as soon as you can ship one. You’ll impress me most if your proposal highlights previous Linux desktop apps you have delivered, especially anything involving biometrics or crypto/fin-tech security.

Deliverables (acceptance criteria)
• Installable .deb or AppImage plus complete source code
• Biometric onboarding & login that functions offline
• Encrypted local wallet capable of deposit, withdrawal, and an interest-style growth simulation
• README with build steps and a quick “getting started” guide

I’m ready to move forward immediately, so if you have solid experience in Linux desktop development and biometric integration, let’s talk right away.

Skills: Linux, User Interface / IA, Cryptography, Software Development, OpenCV, App Development, Software Engineering, Desktop Application

@everyone ┇ BIG REQUEST ┇ NO UPFRONT ┇ PAY ON FULL DELIVERY ONLY

Helios 2 + Titan Two │ Complete Private Discord Dashboard + Full AI Suite + ALL Scripts From Scratch │ NBA 2K26 & Universal FPS

Need ONE real top-tier dev (or small duo) who can deliver the entire package 100% finished. Focus on the best AI tools like YOLO/OpenCV/TensorFlow for god-tier detection, with dead-simple beginner modes (plug-and-play presets, auto-setup wizards, tooltips) and advanced customization (raw variable tweaking, custom model training, script editing IDE in Discord).

Everything included from zero:
- Private Discord bot + pro-level interactive dashboard (embeds that look like Atlas/Phoenix but better – responsive, no lag)
- ALL Titan Two GPC scripts written fresh (anti-recoil patterns, rapid fire mods, perfect release combos, aim injection logic, toggle macros – everything optimized for low detection)
- Helios 2 handling all CV/AI processing
- Full integration: Discord controls Helios → Helios commands Titan Two in real time
- Beginner-friendly: One-tap setups, guided tutorials in bot; Advanced: Unlock Python hooks, threshold sliders, dataset uploads for retraining

Discord Dashboard Details (Scalable UI)
- Clean, modern embeds: sliders for timings/sensitivity, buttons for modes/toggles, dropdowns for profiles/shot types, live graphs for stats (green %, aim drift, recoil curves)
- One-click save/load/share named profiles (export as .gpc or .json)
- Real-time feedback: Detection confidence %, hit/miss logs, visual overlays (preview aim lock or shot meter in Discord)
- Password-locked tester channels + auto-update notifier for new models/scripts
- Beginner mode: Auto-optimize everything with presets; Advanced mode: Full raw access to variables, GPC code editor, AI hyperparams

NBA 2K26 (Current + Next Gen) – Ultimate AI Shot God
- 99-100% green release every single shot with meter AND no-meter modes
- Triple+ detection: Shot meter zones (green/white/yellow), player animations (release point prediction), audio cues (swish/net sounds for feedback), ball trajectory tracking
- Every shot type covered: Jumpers, fades, hops, post spins, stepbacks, dribble pull-ups, catch-and-shoot, off-dribble, leaners, floaters, dunks/layups timing
- Adaptive AI: Detect defender proximity/contest levels (heavy/light/open), auto-adjust release based on stamina/fatigue, badge activation (e.g., Deadeye, Limitless Range)
- Game mode support: MyCareer, Pro-Am, Rec Center, Park – auto-switch profiles based on detected mode
- Patch-proof: One-click retrain button with easy dataset upload (beginners: Bot guides you; pros: Full PyTorch/TensorFlow access for custom models)
- Extra features: Auto-dribble combos for space creation, stamina management (slow down when low), green window expansion via AI prediction
- Zero input lag on HOF difficulty – integrate best tools like PegasusVision CV for pixel-perfect timing

Universal FPS (Warzone, Apex, Valorant, Xdefiant, etc.) – Elite AI Aim Lock Suite
- YOLOv8/v9 (or better) running on Helios 2 for enemy/head/body detection – pre-trained models for instant start, custom training for pros
- All aim modes: Legit (subtle smoothing), Rage (instant snap), Hybrid (toggleable), Silent aim, Triggerbot, Memory/Pixel aim (safest per game)
- Advanced tracking: Wall prediction (track through smoke/walls via movement patterns), aimbot with humanized curves (random offsets, bezier smoothing)
- Live editable recoil tables for every weapon (drag-drop patterns for beginners, script raw values for pros), anti-recoil compensation
- FOV circle customization, priority targeting (closest/low-health/head-first), bone selection (head/neck/chest toggle)
- Extra features: Auto-fire on lock, no-scope assist, quickscope timing, movement prediction for moving targets, anti-ban randomization (variable delays, miss % sliders)
- Stays undetected every season: Auto-update hooks for new anti-cheat patches, mode switcher based on game meta
- Game-specific tweaks: CoD bloom control, Apex legend ability detection, Valorant agent outlines, Xdefiant recoil variance

Tech Requirements (Best-in-Class AI)
- All computer vision/AI on Helios 2 (zero lag, high FPS capture via Elgato/Remote Play)
- You write and deliver ALL Titan Two GPC scripts from scratch (I have nothing – make them modular and easy to flash via Gtuner IV)
- 100% local & encrypted, no cloud dependency – advanced: OpenCV/Python/TensorFlow hooks for custom builds
- Compatibility: PS5/Xbox/PC, future-proof for 2K27+ and new FPS titles

Payment
- ZERO upfront, zero milestones
- Full payment ONLY when I flash the scripts, join the Discord, and everything works perfectly in-game on my console (test beginner + advanced modes)

DM me with:
- Short clip or screenshots of your previous full Helios 2 + Titan Two packages (must show dashboard + working scripts + AI features)
- How long until I get the complete, ready-to-use version
- Your price (paid only on delivery)

New talented devs are 100% welcome — if you’re good with Python/OpenCV, can write clean GPC, and know how to make things easy for beginners while keeping it powerful for pros, hit my DMs. I’ll give you a shot if your proof looks legit.

Let’s build the cleanest, most user-friendly, and strongest 2026 setup ever made.

Skills: Python, User Interface / IA, Software Testing, Machine Learning (ML), Software Development, Computer Vision, Game Testing, Discord API, YOLO, AI Development

We require an expert Mobile Developer (Flutter preferred, or Native) for our shooting range application. The main challenge is precise, real-time visual analysis of targets using the smartphone camera. We are not looking for general mobile developers; expertise in real-time image processing is mandatory.

Critical Technical Requirements (Non-Negotiable)
OpenCV / Image Processing Expert: Must have proven experience integrating and utilizing OpenCV (or similar advanced libraries) for real-time video analysis on mobile devices.

Region of Interest (ROI): The final app must allow the user to define a custom dynamic shape (e.g., a circle or rectangle) on the target, and all shot detection and scoring must occur only within that specific area.

Mandatory Stage 1: Paid Proof of Concept (PoC)
To qualify for the main contract, all candidates must successfully complete this paid screening task:

PoC Task: Deliver a working APK/TestFlight build that successfully detects a single new hole on a paper target in real-time.

Budget for PoC: $200 (Fixed Price).

Deadline for PoC: 3-5 Business Days.

Only developers who successfully complete and demonstrate the PoC will be considered for the full MVP contract.

🎯 The Core User Experience We Seek
The final working application must deliver the following seamless user experience: The user arrives at the range, sets the smartphone camera to focus on the target, and the application must instantly recognize and mark any shots that land on the target area in real-time, providing immediate feedback on hit location and timing.

• Please describe the specific computer vision algorithm you would use to reliably detect a new, small bullet hole on a static paper target in a real-time video stream
• How would you technically implement the Region of Interest (ROI) Masking in OpenCV to ensure the app only detects hits within a custom-drawn circle or shape?
• Which method would you use to integrate OpenCV with Flutter (e.g., platform channel, separate plugin, etc.) to ensure high FPS and prevent mobile overheating?
• What frameworks have you worked with?

We are building an Event Intelligence Platform that connects to existing venue CCTV systems and provides real-time operational insights for events (sports, venues, live events).

We are looking for a Computer Vision Engineer to implement and configure camera-based AI modules, working closely with our Head of R&D.

Scope of work (MVP – 30 days)

You will work on practical, deployable AI, not research.

Primary modules:
• People counting without double-count
• Directional line crossing
• Entry/exit counts at gates
• Zone footfall
• Queue estimation
• Queue length
• Approximate wait time
• Crowd density
• Zone heatmaps
• Threshold alerts
• Basic presence / completion detection
• “Object present / missing” (for install verification)

Technical requirements
• Strong experience with computer vision in production
• Experience with:
• Python
• OpenCV
• YOLO (v5/v8 or similar)
• Object tracking (ByteTrack, DeepSORT, etc.)
• Experience handling video streams (RTSP / CCTV / IP cameras)
• Ability to tune models for real environments (lighting, occlusion, crowds)
• Focus on accuracy + stability, not over-engineering

What you will NOT do
• No facial recognition
• No identity tracking
• No academic research
• No building dashboards (handled by another developer)

Deliverables
• Working AI modules that expose:
• Counts
• Densities
• Queue metrics
• Clear API outputs for integration into our platform
• Configuration instructions for different camera angles

Nice to have
• Retail analytics, smart city, or transport analytics experience
• Experience deploying on edge or lightweight servers

Engagement
• Remote
• Fast-paced MVP
• Immediate start
• High chance of long-term collaboration as platform grows

We are a drone software company with a fully built 2D & 3D web viewer frontend. We are now looking for an expert Computer Vision Engineer to architect a custom, high-performance 3D reconstruction engine to power the backend.
The Goal: We want to build a proprietary processing pipeline that surpasses OpenDroneMap (ODM/WebODM) in quality. While ODM is a great starting point, our goal is "commercial-grade" output: sharper edges on buildings, denser point clouds, and superior texture blending without the "fuzziness" often seen in open-source wrappers.
The Challenge (Backend Only): You will architect a custom pipeline (likely leveraging COLMAP + OpenMVS or similar high-end libraries) to achieve:

Superior Dense Matching: Higher density point clouds than standard OpenSfM.

Engineering-Grade Accuracy: Robust bundle adjustment and camera calibration handling (self-calibration).
High-Quality Texturing: Advanced seam leveling and color correction to avoid texture artifacts.
Performance: GPU-accelerated processing (CUDA) to handle 1000+ image datasets efficiently.

Technical Requirements:

Deep expertise in Photogrammetry: Structure from Motion (SfM) and Multi-View Stereo (MVS).
C++ & Python: Ability to write performant C++ code and wrap it for a Python backend.
Libraries: Extensive experience with COLMAP, OpenMVS, AliceVision, or Ceres Solver.
Math: Understanding of Bundle Adjustment, Camera Intrinsic/Extrinsic calibration, and coordinate system transformations (EPSG).

Nice to Have:
Experience with NeRF (Neural Radiance Fields) or Gaussian Splatting (as a future R&D path).
Experience optimizing large LAS/LAZ files for web streaming (Entwine/Potree).

We are seeking a skilled developer to create Titan Two scripts compatible with computer vision or Helios 2. The ideal candidate should have experience in programming scripts specifically for Titan Two devices and a strong understanding of vision processing systems. Communication will be conducted through Upwork, and we do not accept payments or discussions via Discord. If you have a passion for gaming technology and can deliver high-quality scripts, we encourage you to apply!

**Relevant Skills: **
- Titan Two Scripting
- Computer Vision
- Helios 2 Integration
- Programming (C, C++)
- Problem-Solving
- Attention to Detail
I want to give you an example it a discord name input sense I want to learn this and be successful I just want my own willing to learn and be friends https://discord.gg/6TRJ95k3

• Describe your recent experience with similar projects
• Please list any certifications related to this project

I have more than fifty JPEGs that must be pulled directly from a specific website, saved at their original resolution, and delivered back to me completely free of the site’s watermark. The mark sits in the exact same position on every image, so a repeatable action in Photoshop, GIMP, or a scripted approach in Python/OpenCV should do the trick quickly.

Once all files are downloaded, I need the watermark removed without cropping or blurring the surrounding content. Please keep the colour profile and dimensions intact and return the cleaned images with their original filenames in a single zipped folder. If you automate the task, include the action file or script so I can rerun it later on new images.

Deliverables
• Full-resolution JPEGs, watermark-free, matching the source filenames
• Optional: the batch action or code used for removal

I’ll provide the site URL as soon as we start. A fast turnaround and pixel-perfect results will be highly valued.

Skills: Python, Photoshop, Software Architecture, Photo Editing, GIMP, Image Processing, Image Recognition, Image Analysis

I’m looking for a developer to help finalize a custom React-based dashboard (Material UI v5) that integrates with WooCommerce and an existing Python image-processing API. The core functionality is already built; I need support completing the integration and preparing the entire system for stable production use. I have no time to finalize this.

The work includes improving performance on the WooCommerce side (avoiding inefficient queries such as heavy wc_get_orders usage), optimizing backend logic so the dashboard runs clean and fast without relying on caching, and ensuring the React interface is fully responsive and reliable across devices. You will also help connect the dashboard to the Python API and make general improvements where needed.

The Python script is included in the attachment. If you are not comfortable understanding or optimizing this script, this project is likely not a good fit. Flexibility is important, as I may request adjustments to ensure everything is implemented correctly.

• Tell me how you would approach this. How should the architecture look like for such a project.
• Describe your recent experience with similar projects
• What frameworks have you worked with?

For my academic project I need a small, self-contained Python program that can tell, from a single JPEG image, whether a chair is empty or occupied. The workflow I have in mind is simple:

• I place a JPEG file which have multiple chairs in it (or pass its path as an argument).
• Your script runs and prints, logs, or returns how many chairs are occupied.

The input will always be a static image, not a live feed or video stream. Beyond that, I’m flexible: you can rely on classical OpenCV techniques, a lightweight CNN in TensorFlow or PyTorch, or any hybrid approach you feel offers the best accuracy–as long as the code is clean, well commented, and easy for me to explain in class.

Please include in the hand-over:
1. Source code (.py) with in-line comments.
2. A short README describing environment setup, how to run the script, and how the detection logic works.
3. Optional sample images or links to any open dataset you used for testing, so I can reproduce your results.

Accuracy matters, but clarity and reproducibility matter even more for this assignment. If you think a pre-trained model or transfer learning will speed things up, feel free to recommend it as part of your proposal.

Skills: JavaScript, Python, Software Architecture, Machine Learning (ML), C++ Programming, Artificial Intelligence, Image Processing, OpenCV, Computer Vision, Deep Learning

Overview
We are building a night-vision goggle (NVG) calibration rig using two Raspberry Pi camera modules. The system must overlay or compare the two feeds and produce quantitative scores for both alignment and focus (blurriness). We are only interested in candidates who have already built real image-processing systems on Raspberry Pi using Python and OpenCV. If you have not done work of this nature, do not apply.

Project Requirements
• Capture and process two simultaneous Pi camera feeds
• Perform feature detection and alignment scoring
• Compute a repeatable focus/blurriness metric
• Output numeric alignment and focus scores (0–100 scale preferred)
• Provide basic live display of the results
• Deliver deterministic, stable performance on Raspberry Pi hardware
• Code must be clean, documented, and production-ready

Required Experience (non-negotiable)
• Proven work with Raspberry Pi + Python + OpenCV
• Direct experience with multi-camera input on Raspberry Pi
• Prior implementation of feature matching (ORB/SIFT/etc.)
• Prior implementation of focus or blur metrics (Laplacian variance or similar)
• Ability to demonstrate past projects (GitHub, video, screenshots, explanation)
• Ability to take full ownership of the task without hand-holding

Do Not Apply If
• Your experience is limited to tutorials, school projects, or chatGPT-generated code
• You’ve only used OpenCV on desktop machines
• You have never run two cameras simultaneously on a Pi
• You are “willing to learn on the job” (we’re not paying for your education)

Deliverables
• A complete Python application running on the Pi
• All image-processing logic implemented cleanly and efficiently
• Documentation for setup and operation
• Support during integration and testing with our hardware

Budget and Timeline
Flexible. We prefer someone who can start immediately and finish quickly due to prior experience.

I am looking for an experienced Python + Deep Learning developer to build a complete software system for Emotion Recognition from Handwritten Text.

The system will take an image of handwritten text (Hindi or English), process it using deep-learning models, and detect six human emotions:

Happy

Excited

Tender

Sad

Angry

Scared

This application will be used for an academic research project, so the quality of the model, accuracy, and documentation matter a lot.

Key Features Required
1. Image Upload Interface (GUI) – Tkinter

Upload a handwritten image

Preview inside the app

Predict Emotion button

Train / Re-Train Model button

Display predicted emotion + confidence percentage

Save logs of all predictions

2. Preprocessing Steps

The system should perform essential image processing such as:

Convert to grayscale

Binarization (Otsu or adaptive threshold)

Noise removal

Resizing to model input size

Optional: line/word segmentation for cleaner feature extraction

3. Deep Learning-Based Classification

The model should classify handwritten images into 6 emotions.

You may use:

Model Options

TensorFlow / Keras

A custom CNN

Or Transfer Learning such as:

ResNet50

EfficientNet

VGG16

Output

6 emotion classes

Probability score / confidence for each prediction

4. Handwriting Feature Extraction (Optional but Preferred)

In addition to CNN features, the system should optionally extract handcrafted handwriting features such as:

Ink Density

Slant Angle

Stroke Thickness / Pressure

Baseline Stability

Word Spacing

Line Spacing

If possible, fuse these features with CNN features (feature-level fusion) to improve accuracy.

Deliverables

The final delivery should include:

Fully functional Tkinter GUI Application

Training script + pre-trained deep learning model (.h5)

Custom preprocessing pipeline

Handwriting feature extraction module (if included)

Evaluation report (Accuracy, Precision, Recall, Confusion Matrix)

Clean, modular Python code with comments

README or PDF documentation explaining:

How to run the app

How to add new data

How to retrain the model

Dependencies and environment setup

Tech Stack Required

Python

TensorFlow / Keras

NumPy

OpenCV

Pillow

Tkinter

Scikit-learn

Bonus Skills:

Transfer Learning

Model optimization

Feature fusion (CNN + handcrafted features)

Explainability (Grad-CAM)

I’m looking for a developer to help finalize a custom React-based dashboard (Material UI v5) that integrates with WooCommerce and an existing Python image-processing API. The core functionality is already built; I need support completing the integration and preparing the entire system for stable production use. I have no time to finalize this.

The work includes improving performance on the WooCommerce side (avoiding inefficient queries such as heavy wc_get_orders usage), optimizing backend logic so the dashboard runs clean and fast without relying on caching, and ensuring the React interface is fully responsive and reliable across devices. You will also help connect the dashboard to the Python API and make general improvements where needed.

The Python script is included in the attachment. If you are not comfortable understanding or optimizing this script, this project is likely not a good fit. Flexibility is important, as I may request adjustments to ensure everything is implemented correctly.

• Tell me how you would approach this. How should the architecture look like for such a project.
• Describe your recent experience with similar projects
• What frameworks have you worked with?

Description
I’m looking for a talented full-stack developer to help improve an existing system that analyzes complex PDF documents and converts them into structured data. You’ll work across Python, Node.js, React, and modern AI tools on an already running project.

What You’ll Do

Enhance Python code for processing and analyzing PDF documents

Integrate and refine AI/ML models for document understanding and detection

Build and maintain Node.js REST APIs for the processing pipeline

Work on a React UI to visualize and review extracted data

Debug, optimize, and improve accuracy and performance

Required Skills

4+ years full-stack development experience

Strong Python skills (data processing / automation)

Experience using AI/ML in production (e.g. document/image analysis, model integration, or AI APIs)

Solid Node.js and REST API development

Good React knowledge

Experience working with technical/engineering/architectural PDFs or complex documents

Nice to Have

Experience with CAD, vector, or technical drawing formats

Background in engineering, architecture, or similar fields

Experience with computer vision / document processing frameworks (OpenCV, PyTorch, TensorFlow, etc.)

Offer

Long-term collaboration if it’s a good fit

Clear technical challenges and ownership

Competitive compensation, remote

We are seeking a skilled 3D Computer Vision Expert to develop an efficient solution for 2D image stitching, specifically tailored for measurement applications. The ideal candidate will have a strong background in computer vision techniques and experience working with image processing algorithms. You will be responsible for enhancing our existing systems, ensuring accuracy and reliability in stitching images for precise measurements. If you have a passion for 3D imaging and a proven track record in the field, we would love to hear from you.

Relevant Skills:
- 3D Computer Vision
- Image Processing
- Algorithm Development
- Machine Learning
- Software Development