Parking Space Identification Based On Vision Sensor And Deep Learning Research

With the continuous development of economy and science and technology,the automobile industry gradually prospers and the number of cars increases day by day,which leads to the particular shortage of urban parking space.For most people,it is really a luxury to realize orderly parking in the narrow space,and it may even lead to accidents.Thus,automatic parking technology emerges as The Times require.In this paper,an automatic parking space identification method based on visual sensor and deep learning method is proposed,which can effectively identify the parking space in complex environment in order to provide relevant data basis for the subsequent accurate parking.The specific research contents of this paper are as follows.First,in order to meet the demand of the driver line of sight,and provide the foundation for subsequent parking line detection input,this paper USES the body around four road fisheye camera as a visual sensor,collect images distortion,using Open CV own Kannala distortion projection model for fisheye camera calibration and correction,using inverse perspective transformation algorithm implementation images,the images through a unified perspective transformation of coordinates method to realize four image fusion,and eliminated by the weighted average splicing gap,generate 360 panoramic images.Secondly,the improved YOLOv3 detector was used to detect and locate parking corner points and slot heads,classify parking space types and infer complete parking Spaces.The PS2.0 data set was added with a parking occupancy classification label,and the detected parking Spaces were sent to the improved convolutional network to determine the occupancy classification.Finally,relevant super parameters were set,and the Tensor Flow framework developed by Google brain was used to achieve the above operation.The detection model based on the improved YOLOv3 detector was compared with the parking corner error and detection speed detected based on several other target detection networks.After the parking space category is identified,the improved specific convolutional network is used to classify whether the parking space is occupied.Meanwhile,the performance of the detected parking space occupation classification in terms of classification accuracy,running time,model size and other aspects is compared with that of several existing target detection classification networks.The PSV data set is used to test the performance of the whole model.