Lightweight 3D Multi-Object Tracking via Collaborative Camera and LiDAR Sensors.

Area of Science:

• Robotics and Autonomous Systems
• Computer Vision
• Sensor Fusion

Background:

• 3D multi-object tracking (MOT) is crucial for robotics, autonomous driving, and surveillance.
• Existing 3D MOT methods struggle with false detections, ghost trajectories, incorrect associations, and identity switches.

Purpose of the Study:

• To propose a lightweight and efficient 3D MOT framework using collaborative camera and LiDAR sensors.
• To address key challenges in 3D MOT, including false positives and trajectory management.

Main Methods:

• Confidence Inverse Normalization guided Ghost Trajectories Suppression (CIGTS) module to reduce false detections and ghost trajectories.
• Adaptive Matching Space-driven Lightweight Association (AMSLA) module for efficient and accurate data association.
• Multi-factor Collaborative Perception-based Intelligent Trajectory Management (MFCTM) module for robust trajectory handling.

Main Results:

• The proposed framework achieves state-of-the-art performance on the KITTI dataset.
• Achieved Higher Order Tracking Accuracy (HOTA) scores of 80.13% for Cars and 53.24% for Pedestrians.
• Demonstrated significant improvements in reducing false detections, ghost trajectories, and identity switches.

Conclusions:

• The lightweight 3D MOT framework effectively enhances tracking accuracy and efficiency through sensor fusion.
• The proposed modules successfully tackle critical issues in 3D MOT, offering a robust solution for real-world applications.
• This method provides a promising advancement for perception systems in autonomous vehicles and robotics.