Multi-Object Tracking with TensorRT

🚧 Project Status

This project is under active development and not yet fully complete.

GOAL : This repository provides a modular framework for Multi-Object Tracking (MOT) using various detection models and tracking algorithms in C++ with TensorRT.

---

📂 Project Structure

• Reference

  • Validation scripts directly based on tracking methods from the official GitHub repositories.
  • Analysis of tracking algorithm implementations for easier understanding and further development.

• ONNX_Generator

  • Export PyTorch models to ONNX format
  • Download pre-trained ONNX models

• Detector

  • Build TensorRT engine from ONNX
  • Perform high-speed inference with TensorRT
  • Includes preprocessing and postprocessing modules

• Tracker

  • Implement multiple tracking methods (e.g., ByteTrack, DeepSORT)
  • Assign object IDs and estimate trajectories from detection results

• Run

  • Demo code integrating Detector and Tracker
  • Input → Detection → Tracking → Visualization

---

🚀 Workflow

1. Model Preparation

   • Export PyTorch models to ONNX using ONNX_Generator
   • Or download pre-trained ONNX models

2. TensorRT Engine Build

   • Convert ONNX models into TensorRT engines via Detector
   • Cache and reuse engines for fast startup

3. Inference and Tracking

   • Run detection with Detector
   • Apply tracking algorithms with Tracker
   • Visualize results with integrated demo in Run

---