Research On Parking Lot Vehicle Object Detection Based On Clustering Algorithms

With the increasing popularity of household cars,people often travel by car now.In order to solve the parking problem after arriving at the destination,many car owners will find the location of the nearby parking lot by mobile phone or vehicle navigation.However,such a system can only show the location of the parking lot,and can not provide the owner with parking space information.After entering the parking lot,the owner still needs to spend a lot of time to find the parking space.There is an urgent need for parking lots to provide more detailed information,such as the number of parks available and the number of spare parking spaces.This requires object detection of cars already in parking spaces.This paper combines clustering algorithm with image processing to segment the aerial view image of parking lot and further process the object detection image of vehicles in parking lot,which provides basic information for the subsequent development of a parking lot vehicle information management system.Specific research contents are as follows:(1)Firstly,several common clustering algorithms are introduced: the respective theories of k-means,Mean Shift,FCM and DBSCAN algorithms,and the advantages and disadvantages of each algorithm are discussed.The improved algorithms of k-means algorithm and FCM algorithm are also discussed.The two algorithms and their respective improved algorithms,ISODATA and ENFCM,were applied to the image segmentation,and the parking lot vehicle detection image was obtained after processing such as gaussian filtering,gray level transformation and adaptive binarization.By comparing the detection images of each algorithm,it is proved that the improved algorithm is better than the original algorithm.(2)In this paper,the peak density clustering algorithm(DPC)theory is introduced,and discuss its advantages compared with other algorithms,the proposed to DPC algorithm is applied to image segmentation,under the Lab color space to the color of the image pixel feature vector is extracted,computing the similarity measure between each pixel point,using DPC algorithm for clustering and parking lot vehicle object detection images are obtained.Compared with the object detection image obtained by ISODATA and FCM algorithm,it is proved that the DPC object detection image is not only more robust to image noise,but also more complete compared with the object detection image of the previous two algorithms in vehicle contour.(3))DPC algorithm has many advantages,but there are also some defects: because the number of pixels in the image is generally large,and DPC algorithm is sensitive to the number of data sets,resulting in a long running time.To solve this problem in this paper,the first improvement plan is for vehicle parking lot color image color quantization process,reduce the amount of the color of each pixel in the image,then the screen image after color quantization of DPC clustering process and it is concluded that vehicle object detection image,experiments prove that after dealing with the color quantization of DPC clustering process duration generally shorten by about 40% compared to the original DPC clustering process.The second improvement is the introduction of super-pixel Segmentation Algorithms(SLIC),which is combined with the original DPC algorithm to form a new algorithm SLIC-DPC algorithm.The color and spatial distance values in SLIC algorithm are studied,and the improved distance function is better in processing boundary and preserving shape regularity of super pixel block.The vehicle object detection image obtained by combining DPC and SLIC algorithm is better than the previous ones in terms of vehicle shape retention and noise processing,and the running time is greatly shortened.