Person Detection and Counting System

This project is designed to detect and count people from a live RTSP camera feed. It saves captured frames periodically, processes them to count the number of people detected, and generates reports and visualizations based on the detected data.

Table of Contents

• Features
• Setup
  • Requirements
  • Installation
  • Environment Configuration
• Usage
  • Running the System
  • Understanding the Output
• Files and Directories
• Notes

Features

• Frame Extraction: Periodically captures frames from an RTSP camera feed.
• Person Detection: Processes captured frames to detect the number of people present.
• Real-time Monitoring: Displays the current number of people detected in a live JSON file.
• Daily Reporting: Generates a daily report and graph based on the detections between 08:10 AM and 05:05 PM.
• Threaded Operation: Frame capture and detection run concurrently using Python threads.

Setup

Requirements

The project requires Python 3.10.14 and the following Python libraries:

• opencv-python==4.8.0.76
• supervision==0.22.0
• inference==0.16.0
• python-dotenv==0.21.0
• matplotlib==3.7.1
• numpy==1.24.3

For the extraction module:

• flask
• flask-cors
• opencv-python-headless
• python-dotenv
• requests

Installation

1. Clone the repository:

   git clone https://github.com/sudectn/person-detection.git
   cd person-detection

2. Install the required Python packages:

   pip install -r requirements.txt
   

If using the extraction module, install additional dependencies:

pip install -r extraction/requirements.txt

Environment Configuration

Create a .env file in the root and the extraction directories with the following content:

  CAMERA_USERNAME=your_username
  CAMERA_PASSWORD=your_password
  CAMERA_IP=your_camera_ip
  CAMERA_PORT=554  # Default RTSP port
  CAMERA_PATH=your_camera_path
  ROBOFLOW_API_KEY=your_roboflow_api_key

The ROBOFLOW_API_KEY is required to access the detection model.

Usage

Running the System

The system is divided into two main components: extract_frames.py and detect.py. These can be run separately in different terminals.

1. Running Frame Extraction:

In one terminal, run:

python extract_frames.py

This will start capturing frames from the camera and save them to the image_frames directory.

2. Running Person Detection:

In another terminal, run:

python detect.py

This will process the captured frames, detect the number of people, and update the live JSON files and daily reports.

Understanding the Output

• Live Count: The current number of people detected is saved in output/current_human_count.json.
• Daily Report: A JSON file (output/daily_human_count_report.json) containing the average number of people detected per day.
• Generated Graphs: Graphs showing the average human count per hour are saved in the output directory.

Files and Directories

• extract_frames.py: Captures frames from the RTSP stream.
• detect.py: Processes captured frames and detects the number of people.
• output/: Contains all output files, including JSON reports and graphs.
• .env: Configuration file for camera credentials and settings, including the RoboFlow API key.
• requirements.txt: Lists the Python dependencies.
• extraction/requirements.txt: Lists additional dependencies for the extraction module.

Notes

• Ensure the RTSP camera is properly configured and accessible.
• Run extract_frames.py and detect.py in separate terminals.
• The detection runs between 08:10 AM and 05:05 PM, and the report is generated at the end of the day.
• If you encounter issues with thread synchronization or camera feed access, check the logs for detailed error messages.