Attention-enhanced StrongSORT for robust vehicle tracking in complex environments.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Robotics

Background:

• Multi-object tracking is crucial for autonomous driving but faces challenges like occlusion, scale variation, and feature learning.
• Traditional methods struggle with identity switches, scale sensitivity, and gradient degradation in complex scenarios.

Purpose of the Study:

• To introduce AE-StrongSORT, an attention-enhanced framework designed to overcome limitations in current multi-object tracking algorithms.
• To improve robustness and accuracy in autonomous driving systems under challenging real-world traffic conditions.

Main Methods:

• Developed GAM-YOLO (global attention mechanism-YOLO) for enhanced feature representation and occlusion resistance.
• Introduced F-EIoU loss function to address scale-variant targets and balance learning priorities.
• Optimized CBH-Conv module using Hardswish activation and depthwise separable convolution to mitigate gradient vanishing and maintain efficiency.

Main Results:

• AE-StrongSORT achieved significant improvements on the MOT-16 dataset: 17% MOTA, 2.78% HOTA, and 9.99% IDF1 gains.
• Demonstrated enhanced occlusion-resistant feature representation and reduced identity switches.
• Maintained real-time efficiency with a 17% MOTA improvement at 213 FPS.

Conclusions:

• AE-StrongSORT offers a novel technical pathway for robust vehicle tracking in complex traffic scenarios.
• The proposed innovations effectively address challenges of scale variation, motion blur, and dense occlusion.
• The framework significantly enhances tracking performance, paving the way for safer autonomous driving systems.