Research And Implementation On Vehicle Detection And Tracking System Based On Improved Kalman Filter

Immature behavior,fatigue and careless driving were identified as major safety issues.These human behaviors are considered to be central to the occurrence of traffic accidents.The current response is to equip vehicles with advanced driving assistance systems(ADAS)or unmanned driving systems to reduce the occurrence of traffic accidents,but at the same time this puts forward higher requirements for the development of vehicle technology.The research on the detection and tracking system of the vehicle in front plays a major role in this project.Therefore,a large number of researchers are currently engaged in researches related to vehicle image processing[1]At present,the main problems are included:changing light affects the pixel;The problem of swaying trees shielding vehicles;Problem with bad weather affecting pixels...Under the influence of complex situations,these constantly changing backgrounds make it more difficult to detect and track moving vehicles[2]The main work of this paper is based on previous studies,mainly aiming at the vehicle detection and tracking under different backgrounds.In recent years,with the continuous in-depth study of deep learning,the technology is gradually recognized and used by the majority of scholars.In order to improve the performance of the system,deep learning technology is introduced into the vehicle detection algorithm.The technology has excellent performance in real-time detection of targets of different scales and types in complex scenes[3].In this paper,in order to balance the speed and accuracy of the system,we improve it based on Kalman filter tracking.Due to the large number and variety of targets to be detected in this system,a large amount of calculation is needed in the calculation process.Therefore,we use a simplified version of yolov3 in this system.The convolution layer adopted in this paper is much simpler than yolov3,so that yolov3 does not need to occupy a large amount of memory,thus speeding up the detection speed,but this method will lead to small target detection Missing inspection.In order to solve this problem,we adjust the network structure to meet the needs of the system.In order to enhance the ability of feature extraction and obtain higher detection accuracy,we use k-means clustering to search the model we need on the boundary box of training set more quickly,which makes our data set have better learning ability.In addition,in order to deepen the network layer,a network convolution layer is added to the three original network convolution layers to further improve the performance of yolov3.In today’s background of big data,in order to further improve the performance under different data sets,this paper uses Pascal VOC data set as the basis for filling optimization.The experimental results of target detection show that the accuracy of the improved yolov3 is improved compared with the original framework,which shows that the improved yolov3 has higher application value.In the traditional vehicle tracking,kalman filtering has always been favored by people in order to shorten the error problem during long-term tracking because it can predict the target of the next frame and reduce the probability of drift of the image tracking result of this frame[4].After the detection and prediction results are obtained through Kalman filtering,in order to better distinguish the vehicles with similar appearance,the vehicles with similar appearance not only need to use the area intersection ratio,but also need to use the color histogram of the vehicle for feature extraction.Then,the Hungarian algorithm is used for data correlation to make the tracking more accurate and avoid the problem of missing targets in long-term tracking.In addition,in order to deal with the occlusion problem in the tracking process,this paper proposes a region-based quality assessment network method to reduce the number of tags switching and thus improve the tracking accuracy.Finally,the tracking trajectory is formed after continuous iteration of the system.In order to verify the feasibility of the system,vehicles in different scenes were selected for analysis in this paper.In order to further confirm the superiority of this algorithm,kalman filter algorithm and particle filter algorithm were selected for comparison test in the experiment.The experimental results show that this algorithm is not only more accurate but also more sensitive than the traditional algorithm.