Research On Parking Status Detection Algorithm Based On Image Processing

With the continuous and vigorous development of China's economic society,people's living conditions have undergone earth-shaking changes,and private cars,as a means of transportation,are more and more popular.Therefore,when private cars gradually increased,the traffic problem began to present a austere situation,and became one of the most headache problems to travelers,in which the thorniest obstacle is"parking problem".Consequently,it is urgent to build an intelligent parking management system.Among them,the status information detection of parking spaces is an important part of the system.And it is also the key factor for the subsequent vehicles to enter the parking lot.Hence,it has theoretical and practical significance.To solve above problems,this project studied and summarized the existing domestic and foreign parking status information detection algorithms,and analyzed the disadvantages of those proposed algorithms,and then put forward the following two detection algorithms:parking state information detection algorithm based on areas and parking state information detection algorithm based on voting.These two algorithms were carried out on the basis of obtaining the position information of the empty parking space in advance.Therefore,the algorithm proposed in this paper was mainly divided into two parts:the first part was the position information detection of the empty parking space,and the second part was the status detection of the parking space.In the aspect of location information detection of empty parking spaces,this paper put forward the algorithm of location information detection based on sub-pixel contour.Through experimental research and analysis,the algorithm achieved high accuracy and could detect the position information completely.Detection algorithm based on areas,first used the shape matching template for image registration,and then utilized the location information of empty parking Spaces in the first part to extract the contours in the single parking space respectively.By counting the areas of the contours,the algorithm took the sum of them as the area of a single car,and compared with the corresponding predefined threshold to identify specific state of parking spaces.And detection algorithm based on voting,was mainly to apply image registration to the being tested image at first.And afterwards,the algorithm extracted the entropy,gray contrast and the number of Harris corner of the free parking image and the being tested image as three distinguishing features,and then carried on the comparison of one to one correspondence of the features to distinguish the state information of parking spaces.In this paper,edge detection and closed contour detection were carried out by comparing the multi-feature discrimination algorithm in literature of seventeen.The experimental results showed that edge detection could distinguish the status of the parking space,while the detection effect of closed contour was not ideal.The above two algorithms have achieved better results under a lot of experiments.But for the cases of complex environment and large external disturbance,the test results exist certain deviation.Thus we will plan to make improvements in the later stage in order to improve its stability and practicability,and wish to put into practice in the actual situation.In this paper,the images of outdoor parking lot taken by unmanned aerial vehicle was used as the experimental images.However,the algorithms proposed in this paper was not limited to outdoor parking lot,and it was still applicable to indoor parking lot.In this project,unmanned aerial vehicle of dajiang phantom 4 pro was used as image acquisition hardware equipment to obtain video experimental materials of outdoor parking lot.And the design,implement and test of the program were supported by the HALCON machine vision software developed by Mvtec in Germany.