EnhanceCenter for improving point based tracking and rich feature representation.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Machine Learning

Background:

• Multiple-object tracking (MOT) is crucial for analyzing complex dynamic scenes.
• Existing MOT models often rely on computationally intensive detectors, limiting efficiency.
• Need for robust tracking solutions that maintain performance with reduced computational load.

Purpose of the Study:

• To introduce EnhanceCenter, a novel MOT model.
• To enhance tracking efficiency and stability using a lightweight detector approach.
• To improve performance on challenging MOT benchmarks.

Main Methods:

• Developed EnhanceCenter based on the CenterTrack method.
• Implemented a channel-spatial-spatial feature fusion module for appearance information.
• Optimized backbone network weights for MOT tasks.
• Introduced an improved association method for long-term tracking stability.

Main Results:

• EnhanceCenter demonstrated superior performance compared to models using high-performance detectors.
• Achieved a 1.6% IDF1 improvement and 55.1% HOTA on the MOT17 test set.
• Showcased a significant 13% IDF1 improvement over CenterTrack on the MOT20 dataset.

Conclusions:

• Lightweight detectors can achieve state-of-the-art MOT performance.
• EnhanceCenter offers a more efficient and stable tracking solution for complex environments.
• The proposed methods pave the way for practical MOT applications.