夜间场景中分心驾驶行为的轻量级目标检测与研究OA

Traditional models exhibit significant false positives and negatives when detecting distracted driving behaviors in nighttime environments.To address the issue,this paper develops an improved YOLOv11n optimization model,HCP-YOLOv11n(HGNet CPC PACM).First,YOLOv11n's backbone network is enhanced by incorporating the lightweight neural network architecture PP-HGNet(Paddle Paddle High Performance GPU Net),improving multi-scale feature extraction capabilities while reducing computational burden.Then,CPC(CSP partial convolution)is integrated with the C3k2 model to create C3k2_CPC,which is adopted as the model's head component.The integration enables more efficient extraction of effective features while maintaining low computational overhead.Finally,the PACM(pre-normalization adaptive gating channel enhancement multi-scale dilated)mechanism is proposed specifically for small object detection.By combining multiple dilation rates to expand the receptive field,this mechanism markedly enhances the network's performance in processing small targets and detailed behaviors.Compared to the baseline YOLOv11n model,the improved algorithm improves mAP50 by 3.4％,mAP50-95 by 1.2％,and Recall by 5.8％.Meanwhile,it reduces GFLOPs by 12.7％and parameter count by 24％.Additional experiments on Four Behaviors Dataset further verify the effectiveness of the model,demonstrating its exceptional performances in detecting distracted driving behaviors both in daytime and nighttime.