A Lightweight Attention-Based Deep Learning Framework for Pedestrian Detection in Autonomous Driving Scenarios

Engineering notes

Experimental results demonstrate that the proposed framework achieves a precision of 95.8%, recall of 93.6%, F1-score of 94.7%, mAP@0.5 of 96.4%, and mAP@0.5:0.95 of 82.9%, outperforming the baseline YOLOv8 model and several state-of-the-art pedestrian detection approaches. The results confirm that the integration of attention mechanisms significantly improves pedestrian detection accuracy while preserving computational efficiency.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为端到端自动驾驶、BEV感知、3D目标检测、轨迹预测、路径规划、LiDAR感知等高价值论文补充中文说明。

Original abstract

Pedestrian detection is a fundamental perception task in autonomous driving systems, where accurate identification of vulnerable road users is essential for safe navigation and collision avoidance. However, pedestrian detection remains challenging due to scale variations, occlusions, background clutter, and complex urban environments. This study proposes a lightweight attention-based deep learning framework for pedestrian detection by integrating a Convolutional Block Attention Module (CBAM) into the YOLOv8 architecture. The proposed framework enhances feature representation through channel and spatial attention mechanisms, enabling the network to emphasize pedestrian-related information while suppressing irrelevant background features. The model was evaluated on the caltech pedestrian dataset using standard detection metrics, including precision, recall, F1-score, mAP@0.5, and mAP@0.5:0.95. Experimental results demonstrate that the proposed framework achieves a precision of 95.8%, recall of 93.6%, F1-score of 94.7%, mAP@0.5 of 96.4%, and mAP@0.5:0.95 of 82.9%, outperforming the baseline YOLOv8 model and several state-of-the-art pedestrian detection approaches. Furthermore, the framework maintains a lightweight architecture with only 3.5 million parameters, 9.2 GFLOPs, and an inference speed of 112 FPS, making it suitable for real-time autonomous driving applications. The results confirm that the integration of attention mechanisms significantly improves pedestrian detection accuracy while preserving computational efficiency.

Links and sources

Comments