Research On Detection Method Of Parking Space State Based On Machine Vision

With the rapid development of modern city construction,people’s daily travel mode has become more and more convenient.Many people use private cars for travel,which leads to the aggravation of the bearing burden of public parking lots and the failure to achieve effective parking management.In order to solve this problem,this thesis uses the target detection algorithm of machine vision to identify the cars in the public parking lot,and distinguish the abnormal occupancy of parking spaces such as bicycles,sundries and illegal parking.It realizes the automatic detection of normal parking and abnormal state in the parking lot.It is convenient for the management personnel to manage,timely handle the abnormal situation,and ensure the normal operation of the parking lot.The main contents of this thesis are as follows:(1)Environmental noise preprocessing in image.The environment of natural conditions is changeable,foggy days and insufficient light seriously affect target recognition.In order to obtain clear target feature details,Multi scale Retinex algorithm of color reconstruction based on adaptive color scale and contrast is used for image preprocessing.It effectively reduces the fuzzy interference caused by fog.The S-shaped gain curve is added to equalize the color brightness in the image.The feature details of the image target are complete and the boundary contour is prominent.It is helpful for the algorithm to complete the following target recognition work accurately.(2)Feature extraction of parking space image.The premise of studying the status of parking spaces is to determine the location of parking spaces.Canny operator is used to extract the edge information of parking line,which has a good recognition effect on the contour.Combined with Hough transform algorithm,it has excellent detection effect on spatial line based on parameter space transformation.It can effectively avoid the influence of partial occlusion and discontinuity of parking line.The location of parking space in the image is accurately detected.(3)Construction of algorithm network model.The residual structure is introduced to enhance the depth of the model,enrich the feature information of the image,and improve the recognition accuracy of the algorithm.The size of convolution kernel is increased appropriately to reduce the influence of coherent noise caused by background information.In the detection process,multi-scale candidate boxes are generated in the region of interest to accurately and quickly determine the position of the target in the image.The minimum bounding box is added to the traditional Intersection over Union algorithm.The comparison effect between the predicted target position and the real target position is enhanced.The predicted target position is modified.The border regression is introduced to further optimize the target location.In the process of model training,batch normalization and combined loss function are used to realize the self-correction process of network parameters and improve the recognition accuracy.(4)Parking status detection of parking lot.The preprocessed image is used as the input of parking status detection.The algorithm can detect cars and abnormal status in parking spaces from multiple angles.The results show that,compared with the traditional algorithm,the accuracy is further improved,and the recognition rate is improved under various conditions.It can reduce the rate of missed detection and false detection,enhance the robustness of the system,meet the requirements of practical application,and detect the parking status of parking lot with high precision.