GL-YOLO: Pedestrian Detection Algorithm for Small Objects in Long Distances

A new algorithm named GL-YOLO based on YOLOv8 is proposed to address the issues of low accuracy and high computational complexity in pedestrian detection. Firstly, spatial and channel reconfiguration convolutions are introduced to reduce redundant computations and promote the learning of representative features, thereby reducing the computational load of the model. Secondly, a multi-scale attention module is incorporated to preserve information on each channel while reducing the computational burden, significantly improving the detection accuracy of the model. Finally, an additional small objects detection layer is added, transforming the three detection layers of YOLOv8 into four layers, effectively enhancing the model’s performance in detecting small objects. To validate the effectiveness of GL-YOLO, experiments were conducted on the VOC 2007 and VC2012 datasets. The experimental results demonstrate that compared to YOLOv8, the improved algorithm reduces parameters by 15.3%, increases mAP@0.5 by 4.72%, and mAP@0.5:0.95 by 5.09%. The enhanced algorithm exhibits stronger generalization performance.