Object Detection and Tracking

Overview

This project implements object detection and tracking using YOLOv8 and BoT-SORT/ByteTrack algorithms. The pipeline is designed for processing videos, detecting multiple objects, and tracking them efficiently. It includes preprocessing, hyperparameter tuning, and real-time object tracking. The implementation is done in Python using OpenCV and Ultralytics YOLO.

Features

YOLOv8-based Object Detection: Utilizes the latest Ultralytics YOLO model for high-accuracy object detection.
Multi-Object Tracking: Supports BoT-SORT and ByteTrack tracking algorithms.
Customizable Parameters: Users can configure tracking thresholds, object IDs, and motion compensation settings.
Image Enhancements: Includes blurring techniques for improving detection performance.
Automated Video Processing: Reads input video, applies detection/tracking, and saves the processed output video.
Kalman Filter Adjustments: Provides an option to modify Kalman filter parameters for tracking stability.

Installation

To use this project, install the required dependencies using the following commands:

pip install ultralytics scikit-learn opencv-python

Usage

Run the main pipeline to process videos automatically:

task_pipeline()

Configuration

The tracking pipeline uses configurable YAML files for BoT-SORT and ByteTrack. These files define hyperparameters such as confidence thresholds, tracking buffers, and motion compensation settings.

Sample BoT-SORT Configuration

tracker_type: botsort
track_high_thresh: 0.6
track_low_thresh: 0.3
new_track_thresh: 0.85
track_buffer: 100
match_thresh: 0.6
fuse_score: True
min_box_area: 10
gmc_method: sparseOptFlow
proximity_thresh: 0.6
appearance_thresh: 0.6
with_reid: True

Sample ByteTrack Configuration

tracker_type: bytetrack
track_high_thresh: 0.35
track_low_thresh: 0.025
new_track_thresh: 0.8
track_buffer: 300
match_thresh: 0.8
fuse_score: True
min_box_area: 1