Need AI expert.
Budget: $30 – $250 USD
I need an end-to-end solution that can automatically detect marathon bib numbers from footage captured by two fixed cameras, cut the raw video into individual participant clips, add a preset sponsor logo overlay, and then push a low-resolution copy of each clip directly to the runner via WhatsApp (or a future-proof messaging API) by matching the detected number with a phone record in my database.
Core requirements
• Accurate OCR or vision model tuned specifically for marathon race bib numbers, even under varied lighting and motion.
• Synchronized ingest from two fixed HD cameras; the system must decide which angle to use (or combine) for the cleanest view of each runner.
• Real-time or near-real-time processing so clips are available minutes after the athlete crosses the filming zone.
• Automatic clip generation: each video should start a few seconds before the bib is first visible and end a few seconds after the runner leaves the frame.
• Optional logo overlay on a corner of every clip.
• Database connection (SQL or Firebase preferred, open to alternatives) that links detected bibs to phone numbers.
• Automated WhatsApp dispatch of a compressed, share-ready file together with a customizable text message.
• Simple, browser-based admin dashboard that a non-technical volunteer can use to:
– start/stop recording
– monitor detection accuracy
– resend or delete clips
– upload/edit the bib-to-phone spreadsheet
– swap the overlay logo.
Deliverables
1. Fully working prototype deployed on a cloud VM or local server, ready for race-day use.
2. Source code with clear documentation and installation script.
3. Admin guide (PDF or short video) showing setup, operation, and troubleshooting.
4. One remote handover session for live Q&A.
I’m open to your preferred stack—OpenCV, TensorFlow, PyTorch, FFmpeg, or a commercial vision API—so long as licensing fits an event production environment and the admin UI remains dead-simple. Please outline your proposed architecture, expected accuracy, and any hardware specs I should budget for when you bid.
Core requirements
• Accurate OCR or vision model tuned specifically for marathon race bib numbers, even under varied lighting and motion.
• Synchronized ingest from two fixed HD cameras; the system must decide which angle to use (or combine) for the cleanest view of each runner.
• Real-time or near-real-time processing so clips are available minutes after the athlete crosses the filming zone.
• Automatic clip generation: each video should start a few seconds before the bib is first visible and end a few seconds after the runner leaves the frame.
• Optional logo overlay on a corner of every clip.
• Database connection (SQL or Firebase preferred, open to alternatives) that links detected bibs to phone numbers.
• Automated WhatsApp dispatch of a compressed, share-ready file together with a customizable text message.
• Simple, browser-based admin dashboard that a non-technical volunteer can use to:
– start/stop recording
– monitor detection accuracy
– resend or delete clips
– upload/edit the bib-to-phone spreadsheet
– swap the overlay logo.
Deliverables
1. Fully working prototype deployed on a cloud VM or local server, ready for race-day use.
2. Source code with clear documentation and installation script.
3. Admin guide (PDF or short video) showing setup, operation, and troubleshooting.
4. One remote handover session for live Q&A.
I’m open to your preferred stack—OpenCV, TensorFlow, PyTorch, FFmpeg, or a commercial vision API—so long as licensing fits an event production environment and the admin UI remains dead-simple. Please outline your proposed architecture, expected accuracy, and any hardware specs I should budget for when you bid.
Related categories:
PHP
Python
Software Architecture
Google App Engine
Node.js
Software Development
RESTful API
API Integration