Research On Computer Vision-based Aided Parking Location And Navigation

With the development of society and the increase in the number of cars,it becomes very difficult to find suitable parking spaces in parking lots.In some parking lots,the global satellite navigation system can be used to assist in finding parking spaces,but the accuracy is low.In addition,in indoor parking spaces,the global satellite navigation systems can not be used.Therefore,the research on intelligent automatic parking technology is very important.This paper researches on the key technologies of parking space detection and parking navigation based on computer vision.The specific research content and paper contributions can be summarized as follows:1.Based on the research of scene investigation and visual positioning-related algorithms,an algorithm is proposed for detecting the parking space status and 3-D position,which can be used only by the monocular camera of the parking infrastructure.Firstly,by using parking lot parking lines,this paper uses HSV color space processing,morphological operations and Hough transform line detection to obtain orthogonal groups of lines in 3-D space.Then vanishing point estimation algorithm is used to complete the calibration of the camera and the estimation of distribution algorithm(EDA)is used to calculate the locally optimal camera parameters.The EDA algorithm lifts the constraints of unknown parameters during camera calibration,which can reduce re-projection errors and improve the accuracy of the calibration algorithm.Finally,the visual detection algorithm is researched and used to locate the vehicle position in the parking lot so as to achieve the purpose of detecting free parking spaces in the parking lot and guiding the vehicle to accurately find the parking space.Experimental results on the public dataset prove the feasibility of our proposed algorithm.2.A multi-feature fusion Kalman filter tracking algorithm is proposed,which uses the distance between the 3-D position detected by the vehicle and the planned path line as a feature.Firstly,by integrating multiple cameras in 3-D space,the position information of each camera scene is unified.Then,by combining the color features,size features,and distance feature,an adaptive weight adjustment method is designed to use in non-occlusion scenes,and a method of increasing the distance feature weight is proposed in occlusion scenes.Finally,vehicle tracking is achieved by combining a Kalman filter tracking algorithm.The simulation experiments on the dataset prove that the algorithm effectively improves the accuracy of vehicle tracking in parking lots.