Aiming at the problems that the YOLOv5s algorithm has weak detailed feature learning ability, excessive redundant information, and insufficient key feature fusion in complex highway environments leads to low accuracy of vehicle target detection, a real-time detection model of highway vehicle is proposed based on YOLOv5s (YOLOv5s-CRCP). Firstly, convolutional attention module is embeded in the residual unit to strengthen the learning of detailed features and suppress the interference of redundant information. Secondly, convolutional attention is integrated into pyramid network to distinguish different important information and strengthen the fusion of key features. Experiments are conducted on the constructed Ningxia highway vehicle data set, and the average detection accuracy reaches to 91.2%, which is 4.1% higher than that of original algorithm. Experimental results show that the proposed method has better detection performance in comparison with YOLOv5s and the mainstream real-time vehicle target detection algorithms.