The Research Of Roadside Parking Behavior Recognition Based On Video

With the advancement of urban modernization,the number of private cars has increased rapidly year by year,and the number of parking spaces in cities is seriously insufficient.How to improve the intelligence level of the roadside parking management system,effectively manage and make full use of the roadside parking resources to meet the parking demand has become an urgent problem to be solved.Aiming at this issue,this paper designs a roadside parking timing management scheme based on video,and studies the key,parking behavior recognition technology which identifies the vehicle parking in and out of the parking space,to realize the roadside parking timing function in the scheme.Firstly,the basics of vehicle behavior recognition,moving object detection and multiple object tracking,is studied in depth.Then,based on tracking trajectories,the roadside parking behavior recognition method are studied.The main research work is as follows:1.In order to meet the requirements of real-time and environmental adaptability of moving object detection algorithm in the scene of roadside parking,the Gaussian mixture modeling algorithm is analyzed and improved,and an adaptive Gaussian mixture modeling method combining spatio-temporal information is proposed.Firstly,according to the contradiction between intermittent motion and background updating,a formula for calculating adaptive background learning rate is derived by using the historical matching times and taking the shape of arctangent function curve as the change rule of background learning rate.Then,the foreground pixel discrimination rule is divided into two steps: the preliminary judgment based on three frame difference principle and the judgment based on neighborhoods voting.2.Aiming at the problem that the multiple object tracking method based on detection is too dependent on the object detector,the particle filter Detection-Free tracking algorithm is studied.Under the basic framework of Detection-Based multiple object online tracking algorithm,an improved multiple object tracking strategy based on particle filter is proposed.The joint similarity between objects is calculated by using color features and spatial features,and the data association is performed twice.Based on the results of data association and particle filter prediction and estimation,the trajectories are managed to create,delete and maintain growth.The experiment verifies the validity of the algorithm and can solve the influence of object occlusion and object missing detection on multiple object tracking accuracy to some extent.3.The tracking trajectory is smoothed,and the trajectory states of parking and leaving processes are analyzed.Then,the effective motion parameters,such as speed,acceleration and motion deflection,are extracted to quantitatively analyze the trajectory states.In this way,a roadside parking behavior discrimination model based on trajectory segments is established.The experiment verifies the validity of the model in recognising the roadside parking behavior.The model can effectively recognising the behaviores of start parking,vehicle stopping,start leaving and vehicle leaving during the parking process.The algorithm has good real-time performance and meets the basic requirements of roadside parking timing management.