A lightweight YOLO based framework for real-time seat belt detection in intelligent traffic surveillance

This is the authors' abstract. We don't add key points for this paper.

Road accidents remain a major cause of fatalities worldwide, and proper seat belt usage significantly reduces the risk of severe injuries and death. Ensuring compliance through automated traffic surveillance systems is therefore essential; however, reliable real-time detection remains challenging due to illumination variations, occlusions, windshield reflections, and low contrast between seat belts and vehicle interiors. This paper presents a task-oriented lightweight detection framework, termed as ECA-BiFPN-YOLO, designed specifically for robust seat belt detection under real-world conditions. The proposed approach systematically integrates efficient feature generation, lightweight attention, adaptive multiscale fusion, and visibility enhancement to address the unique challenges of thin-object detection in degraded environments. The framework incorporates GhostC2f modules and depthwise convolutions for computational efficiency, Efficient Channel Attention (ECA) for improved channel-wise feature discrimination, and a Bidirectional Feature Pyramid Network (BiFPN) with learnable weighted fusion for stable multiscale representation. Additionally, a Dark Channel Prior (DCP)-based preprocessing pipeline is employed to enhance visibility in low-light and hazy scenarios. Extensive experiments conducted on the SeatBelts-V1 dataset and a cross-domain dataset demonstrate that the proposed framework achieves an mAP@50 of 95.6% while maintaining real-time inference performance (11.5 ms per image, ~ 85 FPS) on an NVIDIA T4 GPU. Comparative and ablation studies confirm that the proposed design provides an effective balance between accuracy, efficiency, and robustness. These results highlight the practical applicability of the proposed framework for large-scale intelligent traffic surveillance systems, particularly in resource-constrained and real-world deployment scenarios.