Research On Parking Space Detection Algorithms For Indoor Parking Based On Visual Analysis

With the rapid increase in the number of private cars,parking difficulty has become the "pain point" of the city,so how to better solve the problem of parking difficulty has become a research hotspot.At present,besides the insufficient number of parking spaces,another main reason is that drivers do not know the parking space information.For example,they do not know which parking lots are nearby,how many empty parking spaces are available,and where the parking spaces available for parking in the parking lot? If all parking space information can be put on a data platform,the idle state of parking space in a certain area can be known in real time,and the parking resources can be fully allocated and used,which will greatly change the parking difficulty problem.However,if want to obtain parking space information,the first problem to be solved is to automatically detect whether the parking space is free.Some researchers have proposed parking space detection technology based on sensors,ultrasonic waves and infrared rays,but these methods need to excavate the ground,a large number of wiring and other engineering transformations,the construction intensity and cost are quite high,and it is difficult to implement.Because of surveillance cameras are commonly installed in the existing parking lots for security purposes.In this paper,based on the video images collected by these cameras,the image analysis technology is used to detect the parking status of the parking space automatically and timely without obtaining a large increase in engineering construction and cost,so that obtain the parking space idle information of the entire parking lot.The main work of this paper are as follows:1)Through analysis,it is found that when parking spaces are not parked,the feature is mainly gray cement floor or other single-color ground,with less texture information.When there is a vehicle parked,because the body has windows,a license plate,intake grids and so on,the texture is rich,and this feature can be used to effectively detect the idle state of the parking space.2)The collected parking space images are divided into 16?16 blocks,and each block is labeled,with the textured blocks are treated as a positive sample and the untextured blocks are treated as a negative sample.In the experiment,more than 1900 image blocks were labeled,some of which were used as training samples and the other as test samples.3)The energy,contrast,entropy of gray level co-occurrence matrix(GLCM)in the 0°,45°,90°,135° degrees and HOG features of each image block are calculated respectively,and these features are fused to form a feature vector(called GLCM_HOG feature vector in the paper).4)The obtained image block GLCM_HOG feature vector is classified and trained using a Support Vector Machine(SVM)to obtain a parking space image block texture classifier.5)The classifier is used to classify each parking space image block,counting the proportion of texture blocks in each parking space area,calculating the ratio of texture blocks in parking space with car and in idle state,and finding the threshold of the ratio between parking space with car and idle state to determine whether there is a car occupancy in the parking space.