YOLOv10 Object Tracking with DeepSORT

This project demonstrates a complete pipeline for real-time object detection and tracking using YOLOv10 and DeepSORT. It processes video input, detects objects, tracks them across frames, and provides optional blurring for specific object classes.

Features

• Object Detection: Utilizes YOLOv10 for high-accuracy real-time object detection.
• Object Tracking: Employs DeepSORT for robust multi-object tracking.
• Customizable Confidence Threshold: Allows users to set a confidence threshold for object detection.
• Class-Specific Tracking: Enables tracking and counting of specific object classes.
• Gaussian Blur: Optionally applies Gaussian blur to detected objects of a specified class.
• Real-time Performance Logging: Logs processing time for each frame and overall class counts.

Requirements

• Python >= python3.9
• OpenCV
• PyTorch
• NumPy
• absl-py
• deep_sort_realtime
• ultralytics

Installation

1. Make a folder and open an integrated terminal.
2. Clone the Yolov10 repo:
   https://github.com/THU-MIG/yolov10
cd yolov10
pip install .
3. Clone the repository (In the integrated folder you created):
   git clone https://github.com/your_username/yolo_deepsort.git

Usuage

Run the script with the following command:
python track_objects.py --video <path_to_video_or_webcam_index> --output <output_video_path> --conf <confidence_threshold> --blur_id <class_id_to_blur> --class_id <class_id_to_track>

Command-Line Flags

• --video: Path to input video file or webcam index (default: ./data/test_1.mp4).
• --output: Path to save the output video (default: ./output/output.mp4).
• --conf: Confidence threshold for object detection (default: 0.25).
• --blur_id: Class ID to apply Gaussian blur (default: None).
• --class_id: Class ID to track (default: None).

Code Overview

Initialization

• Logging Setup: Configures logging for the script.
• Command-Line Flags: Defines flags for video input, output, confidence threshold, and specific class IDs for blurring and tracking.

Functions

• initialize_video_capture: Initializes video capture from file or webcam.
• initialize_model: Loads YOLOv10 model and sets the appropriate device (CPU/GPU).
• load_class_names: Loads class names from a file.
• process_frame: Processes each frame to detect objects using YOLOv10 and tracks them with DeepSORT.
• draw_tracks: Draws bounding boxes and labels on the frame, and applies Gaussian blur if specified.

Main Logic

• main: The main function that ties everything together. It initializes video capture, model, and other resources, processes each frame for object detection and tracking, and handles saving the output video.

Logging and Output

• Frame Processing Time: Logs the time taken to process each frame.
• Class Counts: Logs the counts of each detected and tracked class at the end of the video.