An Improved Lightweight Dense Pedestrian Detection Algorithm Based on YOLOv8

Pedestrian detection is very important for computer vision and self-driving cars. The use of advanced detection techniques such as YOLOv8 in autonomous driving suffers from high algorithmic complexity and a substantial number of computational parameters. As such, we propose a novel lightweight pedestrian detection approach based on an enhanced YOLOv8 model, namely the YOLOv8-GS network. The proposed C2f_GhostDynamicconv module in our network integrates more efficient Dynamic convolution and Ghost Module, and some Conv blocks in the Backbone and Neck sections are replaced by the more lightweight GSConv blocks, thus reducing the number of parameters and improving the detection speed without affecting the detection accuracy. Evaluation on the open WiderPerson dataset reveals that YOLOv8-GS has the same accuracy as YOLOv8, but with 39.1 % fewer parameters and 66.3 % faster computation, making it more suitable for lightweight pedestrian recognition.