Retail People Tracking & Analytics
Budget: ₹600 – ₹1,500 INR
I need a complete Python pipeline that ingests CCTV streams from several in-store cameras and turns them into actionable retail metrics. Using YOLOv8 for detection, ByteTracker plus a solid Re-ID module for cross-camera identity preservation, the system must
• detect and track people only (no vehicles, animals, etc.)
• avoid double-counting the same shopper when they move between overlapping views
• distinguish staff from shoppers so employees are never included in customer statistics
• produce reliable foot-traffic counts, crowd-density heatmaps and an “entry-to-purchase” conversion figure (how many visitors become paying customers)
Please deliver:
1. Well-structured, documented Python code (YOLOv8 + ByteTracker + ReID) that runs on recorded footage or live RTSP streams.
2. A small demo dataset or clear instructions so I can replicate results on my own cameras.
3. Output examples: CSV or JSON counts per time slice, density visualisations and optional dashboard/notebook for quick inspection.
4. Brief performance report showing tracking accuracy and how staff filtering was achieved.
Accuracy, clean reproducible setup, and clear documentation will be the main acceptance criteria.
• detect and track people only (no vehicles, animals, etc.)
• avoid double-counting the same shopper when they move between overlapping views
• distinguish staff from shoppers so employees are never included in customer statistics
• produce reliable foot-traffic counts, crowd-density heatmaps and an “entry-to-purchase” conversion figure (how many visitors become paying customers)
Please deliver:
1. Well-structured, documented Python code (YOLOv8 + ByteTracker + ReID) that runs on recorded footage or live RTSP streams.
2. A small demo dataset or clear instructions so I can replicate results on my own cameras.
3. Output examples: CSV or JSON counts per time slice, density visualisations and optional dashboard/notebook for quick inspection.
4. Brief performance report showing tracking accuracy and how staff filtering was achieved.
Accuracy, clean reproducible setup, and clear documentation will be the main acceptance criteria.