Research And Application Of Robust Vehicle Detection And Tracking Method Based On Deep Learning

Real-time and accurate detection and tracking of the vehicles in the video,so as to obtain the relevant information of the vehicles,can not only provide important information for intelligent traffic control,but also have important significance for assisted driving and automatic driving.However,in a complex driving environment,the interference of factors such as the small proportion of the relative image area of the vehicle and the mutual occlusion will seriously affect the performance of detection and tracking.Therefore,in order to achieve real-time robust detection and tracking of vehicles in complex traffic backgrounds,a vehicle detection method based on multi-scale fusion technology and attention mechanism and a vehicle tracking method combined with kalman filter and dynamic convolution technology are proposed.The main research contents of this paper are as follows:1.Aiming at the phenomenon that the existing Single Shot Multi Box Detector method has insufficient detection accuracy for small targets and serious missed detection,a small target vehicle detection algorithm based on multi-scale fusion technology and attention mechanism is proposed.In order to improve the effect of feature extraction,the effectiveness of the algorithm is verified by experiments.2.Aiming at the occlusion phenomenon between vehicles and the requirement of real-time tracking in the complex traffic background,a lightweight network tracking method based on Kalman filter and dynamic convolution is proposed.The information is matched with the detection results to achieve real-time robust tracking of vehicles,and the performance of this tracking method is evaluated using the public vehicle dataset UA-DETRAC.3.Using the Python programming language,combined with the crossplatform Py Qt toolkit of GUI applications,the method proposed in this paper is deployed to the Windows platform to build a vehicle detection and tracking system for demonstration.In addition,by labeling the surveillance video,a vehicle test set is constructed,and the robustness of the algorithm in this paper under different weather conditions is verified from three scenarios: sunny day,night,and rainy day.