Research On High-Altitude Video Based Roadside Parking Detection Algorithm

With the continuous development of cities and the improvement of people’s living standards,the increasing number of vehicles has brought great pressure to the management of roadside parking lots.Studying roadside parking management systems is an effective way to solve the above problems,among which roadside parking detection technology is the key to the system.Roadside parking detection technology based on high-altitude video has the advantages of high detection efficiency and strong anti-interference ability,but there are still some problems with this technology.First,the parking space lines in the high-altitude video are blurred and the shooting angle is steep,which increases the difficulty of parking space recognition,resulting in low recall and precision rates of parking recognition algorithms and poor matching degree between parking space recognition box and parking space.Second,the network structure of the deep-learning-based vehicle detection algorithm is complex,with a large number of parameters and computation,making it difficult to deploy the algorithm on low-performance devices such as the high-altitude cameras.Therefore,a lightweight and efficient vehicle detection algorithm is necessary.Third,vehicles in the high-altitude video may be blocked,leading to the loss of vehicles tracked by the algorithm.In response to the above problems,this dissertation conducts the following research:(1)A parking space recognition algorithm based on line detection and density clustering is proposed.Firstly,in order to filter out most of the interference in the scene,the algorithm preprocesses the input of high-altitude image.Secondly,the algorithm detects the lines in the image and filters them through density clustering algorithms to obtain the parking space lines.Finally,the algorithm extracts the region of interest containing parking space in the image based on the parking space lines and locates the parking space vertices based on the position relationship between the straight lines,line intersections,and corners in the area,thus determining the position of the parking space.Experimental results show that,compared with existing parking space recognition algorithms,the proposed algorithm has significantly improved recognition performance,with a recall rate of 94.84% and a precision rate of97.83%,and the matching degree between the parking recognition box and parking space is higher.(2)A vehicle detection algorithm based on improved YOLOv5 is proposed.In order to reduce the number of network parameters and computation of the YOLOv5 algorithm,the Ghost Net is used to replace the CSPDarknet53,which is the backbone network of YOLOv5.In addition,the original CIo U loss function of the network is replaced by the EIo U loss function to improve the convergence speed of the network.Experimental results show that compared with the original YOLOv5 algorithm,the improved algorithm reduces the network parameter count by 41.3%,the computational complexity by 47.5%,and increases the number of images processed per second by 18 frames,with a recall rate and precision rate improved by 2.07% and 0.75%,respectively.(3)A vehicle tracking algorithm based on improved Strong SORT is proposed.During the process of tracking vehicles in the high-altitude video,it is easy to lose track of the vehicles due to occlusion.In response to this problem,an RGA attention mechanism module is added to the Res Net50 feature extraction network of the Strong SORT tracking algorithm,which enhances the network’s attention to vehicle features.Meanwhile,the improved YOLOv5 vehicle detection algorithm is used as the vehicle detection algorithm of Strong SORT to improve the vehicle detection performance.Experimental results show that compared with the original Strong SORT algorithm,the improved algorithm achieves a3.17% increase in Multiple Object Tracking Accuracy(MOTA),a 2.63% increase in IDF1,and the algorithm’s processing speed also increased by 3 frames per second(FPS).(4)Based on the proposed parking space recognition algorithm,vehicle detection algorithm,and vehicle tracking algorithm,a roadside parking detection system based on the high-altitude video is designed and implemented.The main function of this system is to automatically identify the parking space status and the vehicle occupying the parking space under the high-altitude video.We collected the high-altitude videos that include footage of vehicles entering and exiting roadside parking spaces,and use them to test the performance of each functional module of the system.The results showed that the system’s success rate in identifying parking space status was 93.78%,and the success rate in identifying occupying vehicles was 96.52%.