Customer Time Tracking in Store

Overview

This project implements a real-time customer tracking system using computer vision to monitor how long customers spend in a store. It combines YOLO v11m for person detection with BoT-SORT tracking to maintain consistent customer IDs and measure their duration of stay.

Getting Started

To get started with this project, follow these steps:

Prerequisites

1. Clone the repository:

   git clone https://github.com/yourusername/customer-time-tracking.git
   cd customer-time-tracking

2. Install dependencies:

   pip install -r requirements.txt

Configuration

1. Create a config.yaml file with your camera settings
2. Adjust detection parameters in settings.py
3. Place your video files in the streams/ directory

Running the System

1. For live camera feed:

   python time_notebook.py --source 0

2. For video file:

   python time_notebook.py --source streams/your_video.mp4

Key Components

• YOLO v11m: Used for person detection
• BoT-SORT: Deep learning-based tracking algorithm
• OpenCV: For video processing and visualization
• CUDA: GPU acceleration support

Features

• Real-time person detection
• Persistent ID tracking across frames
• Time measurement for each detected person
• Visual display of tracking information
• Video output capability

Technical Implementation

1. Initialization

   • Sets up YOLO model with CUDA support if available
   • Initializes BoT-SORT tracker
   • Opens video source

2. Main Processing Loop

   • Reads video frames
   • Performs person detection using YOLO
   • Updates tracker with detection results
   • Maintains time dictionary for each tracked ID
   • Displays bounding boxes with IDs and time information

3. Time Tracking

   if track_id not in tracker_time:
        tracker_time[track_id] = [perf_counter(), perf_counter()]
   tracker_time[track_id][1] = perf_counter()

Requirements

• Python 3.x
• OpenCV
• Ultralytics YOLO
• CUDA-capable GPU (recommended)
• Deep SORT implementation
• Required Python packages:
  • opencv-python
  • ultralytics
  • torch
  • numpy
  • easydict

Usage

1. Ensure all dependencies are installed
2. Place your input video in the streams directory
3. Run the script:

   python time_notebook.py

Output

• Real-time visualization with:
  • Bounding boxes around detected persons
  • Tracking IDs
  • Time duration for each person
• Option to save processed video

Limitations

• Requires good lighting conditions
• Performance depends on GPU capability
• May lose tracking in crowded scenes

Future Improvements

• Add multiple camera support
• Implement zone-based analytics
• Add customer flow patterns analysis
• Export tracking data to database