Pedestrian Object Detection Algorithm Based on Improved YOLOv5s

Traditional pedestrian object detection techniques, represented by YOLOv5, are limited in real-world applications due to their large parameter volumes and high computational complexity. In response to this problem, we propose an improved YOLOv5s pedestrian object detection algorithm. Firstly, we introduce the lightweight network GhostNet into the structure of YOLOv5s to achieve lightweight detection networks. On this basis, a CBAM based on the attention mechanism is proposed to enhance feature extraction and improve the performance of the model while reducing the model parameters. Secondly, we replace the CIoU loss function in the original network with the EIoU loss function, which enables the network training process to converge faster. Finally, we construct a pedestrian object detection dataset to verify the accuracy and real-time performance of the model. The experimental results show that while maintaining high accuracy, the parameter volume of the improved YOLOv5s algorithm is reduced by 45.8% compared to the original algorithm.