Research on pedestrian detection algorithm based on low light environment

Abstract Pedestrian detection in low-light environments often faces many problems such as low detection accuracy, poor real-time performance, and missed detection caused by image blurring. In this regard, an improved algorithm SGF-YOLO(Squeeze and Excitation- GSConv- C2f-Faster) based on YOLOv8n is proposed. By adding a small target detection layer, the missed detection rate of the network in a complex environment is reduced ; secondly, the SE (Squeeze and Excitation Networks) attention module is introduced into the backbone network to improve the model detection accuracy without increasing the parameters of the model. Then, the standard convolution of the deep part of the backbone network is replaced by the lightweight convolution module GSConv ; FasterBlock is used to reconstruct the C2f module to form a new lightweight module C2f-Faster, which not only improves the feature extraction ability of the model, but also reduces the computational overhead. The experimental results show that the improved SGF-YOLO algorithm has an increase in accuracy (P) and recall(R) compared with the benchmark model, and the [email protected] and [email protected]:0.95 are increased by 2.1% and 4.2%, respectively.It can be shown that the SGF-YOLO algorithm model can meet the high real-time requirements in low light environment.