An urban road vehicle detection method based on swin transformer

Vehicle detection is a research priority in the field of transportation. How to quickly and accurately identify the category of the target to meet the real-time and high-precision requirements of vehicle detection is a problem that needs to be solved urgently. Therefore, a Swin Transformer-based urban road vehicle detection model is designed, which takes Swin Transformer as the basic framework and optimizes it from three aspects: the backbone network, the neck network, and the loss function. We construct a dataset of urban road vehicles containing 45,712 images. The experimental results show that the Precision of this model is 93.2%, Recall is 88.3%, and mAP is 92.5%. Compared with Faster R-CNN, SSD and CenterNet models, the Precision is improved by 6.5%, 5%, 2.8%, the Recall is improved by 5.8%, 4.2%, 2.9%, and the mAP is improved by 7.3%, 5.8%, and 4.4%, respectively, with better detection ability. And the AP and mAP values in the car, bus, and truck classes are higher than those of other models. The results show that the model has good accuracy and real-time performance, which can effectively improve the efficiency of traffic management and contribute to the development of traffic informatization and intelligence.