Research On Vehicle Tracking Methods Based On Correlation Filters

In recent years,with the development of artificial intelligence,information technology in transportation has brought a lot of convenience to people’s travel,such as intelligent road traffic management,intelligent transportation and vehicle-assisted driving.However,with the frequent occurrence of intelligent car accidents in recent years,people have begun to worry about the safety and reliability of transportation.This indicates that there is still room for improvement in these technologies.Object tracking is a visual technology that predicts the motion state of an object and labels its position,and it provides information processing support for the aforementioned information technology in transportation.Currently,there are numerous theories and algorithms for object tracking.However,in practical transportation scenarios,tracking performance is reduced when encountering target occlusion,motion blur,and complex backgrounds.This article further studies vehicle video object tracking in response to the above situations.To address the issues of cluttered backgrounds,motion blur causing tracking box drift,and reduced accuracy in vehicle tracking,this paper proposes a BASRCF tracking algorithm based on spatial regularization of correlation filters.First,a background-aware algorithm using real background samples is constructed.Then,starting from its objective function,spatial regularization is introduced to adaptively combine background information and spatial information,followed by optimization and solution of the filter to calculate the target tracking result.The algorithm uses an adaptive spatial weight regularization term to enhance the ability to distinguish between the foreground of the target vehicle and the background.To address the problem of changing target size during vehicle movement affecting tracking accuracy,this paper trains two filters separately: one responsible for predicting the center position of the target vehicle and the other specifically for predicting the size of the target vehicle.Through comparative experiments on the OTB100 and UAV123 datasets,the improvements in spatial regularization and the addition of scale filters have been shown to increase the accuracy of the tracking algorithm.However,when the target vehicle is occluded,the features within the target box cannot fully express the target vehicle,which can lead to tracking failure when predicting the target position based on the target vehicle’s feature information.To address the issue of tracking failure and reduced accuracy caused by occluded target vehicles during tracking,this paper proposes a BASRCF＿RE tracking algorithm that incorporates Kalman filter with correlation filters.First,an occlusion detection mechanism is established by using the peak side lobe ratio(PSR)response graph to determine the degree of occlusion.Then,through Kalman filter motion estimation,the new position of the target vehicle is obtained to perform the tracking task after occlusion.Comparative experiments were conducted on occlusion-related video datasets in OTB100 and UAV123,and according to the experimental tracking results,from car tracking effect diagrams,algorithm success rate,accuracy and other aspects of the analysis,it was found that the improved BASRCF＿RE algorithm proposed in this paper can adapt well to occlusion conditions during vehicle movement,achieving stable tracking performance.