Real-Time Object Detection and Tracking for Retail Analytics

Detects and tracks customers/products from live camera feeds to generate retail footfall and shelf-interaction analytics.

Shop owners know how many people buy, but not how many walk in, where they walk, what they look at or which shelves they touch. Camera-based analytics can answer those questions without any change to the customer's behaviour. This project builds a real-time system that detects and tracks people and products in a live camera feed and turns the movements into retail analytics.

A YOLO object detection model finds customers and products in each frame, and a multi-object tracker assigns each person a persistent identity across frames so that one person is counted once, even when they are briefly hidden. The store floor is divided into zones, and the system counts footfall, measures dwell time and detects shelf interactions. The results are turned into heat maps showing where people spend time. OpenCV handles video capture and drawing, and a Flask dashboard shows live counts, heat maps and hourly trends. The project discusses accuracy, performance on ordinary hardware and privacy, including processing video without storing faces.

You will learn object detection, tracking, real-time video processing and dashboard design. The Project Reference Guide documents the pipeline and evaluation, and the Reference Implementation provides the detection and tracking code, the dashboard and clear set-up steps.