Aiming at the problems of low detection accuracy and low detection speed of the current environment perception net-work model,a multi-task traffic perception network was proposed,in which traffic objects,lane line segmentation and driving area segmentation were simultaneously detected in a more effective way.Using a more powerful and efficient network for feature extraction was conducive to the fusion of richer feature information,so that the detection head and segmentation head had better expression effects.A more effective loss function was proposed,and the direction matching between the real box and the predic-ted box was fully considered in the bounding box loss to improve the training speed and inference accuracy of the model.The coordinate attention mechanism was adopted in the segmentation branch.By adding position information to the channel attention,the ability of the network to perceive the shallow information of the feature map was enhanced,which helped the segmentation head to better identify targets.The model was tested on the BDD100K dataset.The results show that the detection accuracy and inference speed of the model achieve better results.
维普
万方数据