Research On Parking Detection Algorithm For Surveillance Video

With the rapid development of information technology,the intelligent transportation system is continuously innovated.The intelligent transportation system can reduce human resource and recognize some abnormal incidents automatically by using the information technology.Nowadays,the problem of urban traffic congestion is becoming more and more seriously,the traditional solution to avoid those incidents is to install numerous intelligent surveillance devices in public places.However,a minority of drivers still park their cars in the area where parking is prohibited,making the traffic more crowded.Intelligent parking detection technology is one of the important researching tasks of computer vision technology applied in the intelligent transportation system field and possesses profound significance.But in the current monitoring system,twenty-four hours online supervisory control is performed by manual operation,and it wastes manpower and financial resources.As a result,the study of intelligent parking detection is significant.This paper is mainly to study the parking detection algorithms facing the surveillance videos in the case of outdoor complex environment condition s.But there are many defects in the current algorithms: the current algorithms do not consider the impact of shadow of the foreground objects;they can’t solve the temporary parking problem in the process of detection of static objects very well;they only use simple geometric characteristics to determine whether it is a car,resulting in low accuracy of vehicle detection.This paper includes detail introductions of theoretical knowledge and technology difficulties involved in the parking detection based on the surveillance videos,and improves the current parking detection algorithm.The parking detection algorithm in this paper is divided into five main modules: the extraction of foreground moving target,shadow detection,static target detection,vehicle detection and occlusion detection.In the foreground moving object extraction process,the foreground moving objects is obtained by using the self-adaptive hybrid Gauss model.In the shadow detection process,HSV color space model is used to remove the influence of shadow of the foreground object.In the static target detection process,the method of detecting some suspicious stationary targets based on the historical foreground pixel feature detection is put forward to extract the suspicious static areas,we can use the improved hash perception algorithm to judge whether the contents in two pictures are consistent after N frames,and confirm the result whether the target is a static one.Compared with the traditional method,the stationary target detection method in this paper can solve the vehicle retention situation.As for the vehicle detection,the improved cascade Haar classifier is used to detect whether the stationary target is a vehicle,this method can improve the accuracy of vehicle detection.In order to improve the detection rate furthermore,the occlusion detection method based on mixed Gauss model is proposed in this paper.Experiments show that the detection rate of the algorithm on i-LIDS data set is 95.95%.