The Research On Pedestrian And Vehicle Detection Algorithm Based On Improved YOLOX-L

In recent years,pedestrian and vehicle detection algorithms based on convolutional neural networks have achieved rapid development.Pedestrians and vehicles,as two types of important detection targets in road traffic scenarios,are information sources of control and decision-making behaviors during driving.In order to improve the detection effect of current pedestrian and vehicle detection algorithms,this article researches and improves based on the YOLOX-L pedestrian and vehicle detection algorithm.The main works are as follows:(1)The Swin Transformer network is used to improve the backbone network of the YOLOX-L model,the CBAM attention module is added to the feature enhancement part of the model,and the Focal Loss function is used to replace the confidence loss function BCE Loss of the YOLOX-L model to obtain an improved Swin-YOLOX-L model.(2)The lightweight network Efficient Net-B2 is used to improve the backbone network of the YOLOX-L model,the multi-scale feature fusion technology is used to increase a shallow feature layer,and the depth separable convolution,CBAM module and Focal Loss function are introduced into the model to obtain an improved Efficient-YOLOX-L model.(3)The two improved models are compared with the original YOLOX-L model.The results show that compared with the YOLOX-L model,the detection accuracy of the Swin-YOLOX-L model is improved by 0.63%,the model has fewer missed detections and false detections,and has better adaptability to changes in light and weather.The number of parameters of the Efficient-YOLOX-L model is reduced by51.21%,the detection speed is increased by 13.63%,and the detection accuracy reaches 88.94%.It is more suitable for deployment in embedded mobile visual terminals with limited computing power and memory resources.