To address the problem that the vehicle target detection technology of traffic monitoring videos has high rates of false detection and missed detection due to serious vehicle occlusion in traffic congestion periods such as morning and evening peaks, an improved vehicle target detection model based on YOLOv5s network was proposed. The attention mechanism SE module was introduced into the Backbone network, Neck network layer, and Head output of YOLOv5s, respectively, thus enhancing the important features of the vehicle and suppressing the general features. In doing so, the recognition capability of the detection network for the vehicle target was strengthened, and training and tests were performed on the public data set UA-DETRAC and self-built data set. The results show that the three indicators were significantly enhanced compared with the original network, which was suitable for the introduction of the attention mechanism. The evaluation rate, the value, and mean average accuracy were evaluated, and the results showed that compared with the original network, the three indicators were significantly improved, suitable for the introduction of attention mechanisms. To address the imbalance between positive and negative samples and that between difficult and easy samples in YOLOv5s network, the network combined the focus loss function Focal loss and introduced two super-parameters to control the weight of unbalanced samples. Combined with the improvement of attention mechanism SE module and focus loss function, the overall performance of the detection network was improved, and the average accuracy was improved by 2.2 percentage points, which effectively improves the index of false detection and missed detection in the case of large traffic flow.