Personnel Detection Algorithm Based on Spatiotemporal Mapping Event Frames

The personnel detection algorithms based on frame-based traditional cameras exhibit low accuracy in harsh lighting conditions or rapidly moving scenes. To address this issue, we propose a personnel detection network called YOLO-SMCA, which utilizes an event camera to capture personnel motion information based on Spatiotemporal Mapping Event Frames. By incorporating temporal information into the event mapping method to generate event frames, we overcome the limitations of traditional event representation methods, which often lead to the loss of temporal information. This enhancement significantly improves the feature representation capability of the input events. Building on this foundation, we employ the YOLOv5n neural network for personnel detection using spatiotemporal mapping event frames. Additionally, we embed the Coordinate Attention (CA) mechanism in the head of the network to enhance the model’s focus on key features during final decision-making. Through comparative and visualization experiments conducted on the PEDRo dataset, the YOLO-SMCA achieved a mean Average Precision (mAP) of 98.1%, outperforming other methods and meeting the accuracy requirements for personnel detection in complex environments.