Abstract:Highway passenger and cargo mixed operation is an important incentive for traffic accidents. According to statistics, more than 10 people died in traffic accidents, and the collision between trucks and buses accounted for 75%. Although both home and abroad have begun to promote the passenger and freight division, but the supervision of the passenger and freight traffic is not perfect, the current supervision means is also mainly camera photography, later by the staff on the violation of the inspection, and the lack of a mature intelligent detection scheme about the passenger and cargo separation. A video image detection method is used to study and apply to the expressway. In order to improve the accuracy and stability of video detection, a video image detection model based on computer vision and depth learning is built in this paper. The Scale invariant feature transformation (SIFT) pool based on the scale invariant feature transform is proposed. The model of vehicle feature extraction was used to remove the shortcomings of traditional video background modeling with low stability and accuracy, and the feature data and running parameters of vehicles were obtained. After the pilot sample training, the experimental results show that the accuracy of vehicle recognition is as high as 95%, and the accuracy of vehicle lane detection can also reach about 90%.