This repository provides a re-implementation of MOTDT, a real-time multiple object tracking system designed for people. It addresses the challenges of robustly tracking individuals in crowded scenes by combining deep learning-based candidate selection with person re-identification. The target audience includes researchers and practitioners in computer vision and surveillance.

How It Works

MOTDT employs a two-stage approach. First, it uses a deep learning model to efficiently select candidate bounding boxes likely to contain people. Second, it leverages person re-identification features to associate these candidates across frames, maintaining identity consistency. This method aims for real-time performance while achieving high tracking accuracy.

Quick Start & Requirements

• Install dependencies: pip install -r requirements.txt
• Build: sh make.sh
• Run evaluation: python eval_mot.py
• Requires MOT16 dataset and pre-trained weights (links provided in README).
• Official evaluation devkit available at https://bitbucket.org/amilan/motchallenge-devkit/.

Highlighted Details

• Achieves IDF1 scores ranging from 37.1% to 61.1% on MOT16 validation sequences.
• Claims real-time performance, though specific FPS is not detailed.
• Integrates person re-identification for improved identity persistence.
• Provides results from both official devkit and a Python evaluation script.

Maintenance & Community

• The project is authored by Chen Long, with contributions from Haizhou Ai, Zijie Zhuang, and Chong Shang.
• No explicit community channels (Discord, Slack) or roadmap are mentioned.

Licensing & Compatibility

• The README does not explicitly state a license. The code is provided for research purposes.

Limitations & Caveats

• This is a re-implementation, and results may differ slightly from the original paper.
• Performance is evaluated on MOT16 validation sequences; test set results are linked but not directly reproducible from the repo without further steps.
• The absence of a clear license may pose restrictions for commercial use.