Parking space detection method and device

Abstract

A parking space detection method. Aiming at the complex environment of automatic parking system parking space detection, the supervised learning method is used to detect the position and category of candidate parking space corners, and then the detected parking spaces are detected using strong parking space line visual constraints. Secondary confirmation of corner points and categories; coupling vehicle track information to realize matching of corner points of parking spaces based on greedy search; combined with geometric attributes of parking spaces, updating and tracking of corner points of parking spaces and pairing of corner points of parking spaces in time-space sequence Operation; High-precision automatic parking space detection is realized, which greatly improves the success rate of autonomous parking of actual self-driving vehicles.

Description

technical field [0001] The present disclosure relates to the field of visual parking space detection, in particular to a corner-scanning visual parking space detection method, especially to a corner-scanning visual automatic parking space detection method, device, electronic equipment and storage media. Background technique [0002] As the number of cars continues to increase, people's demands for automated and convenient transportation tools are becoming stronger and stronger. With the rapid development of automatic driving technology, the demand for automatic parking is also increasing. [0003] Automated parking refers to the ability of a vehicle to park itself without human intervention. As a key technology in the field of intelligent driving, its research interest is increasing year by year, and parking space perception is a key link in automatic parking. [0004] At present, the parking space perception technology mainly involves the spatial parking space perception ...

AI Technical Summary

Problems solved by technology

[0011] First, the constraints proposed by the current method of extracting the visual attributes of the parking space line according to the feature line segment of the bright line are still slack in the detection of the bright line of the parking space line, and it is difficult to filter the line outline of the shadow strip

[0012] Second, the method of using vehicle dead position information to track the area of ​​interest of the parking space, the cumulative error of dead reckoning needs to be eliminated by "reset operation", which will inevitably increase the overall complexity of the system, which is not conducive to the low-level operation of the actual self-driving vehicle. It is difficult to meet the real-time control requirements of the actual self-driving vehicle when it is automatically parked.

[0013] Third, the commonly used method based on machine learning detects the corner position and type of parking spaces. Generally, it is necessary to calculate the angle between parking spaces according to different types of parking spaces, and then calculate the corner points of the two parking spaces at the far end; but the angle between parking spaces for inclined parking spaces It needs to be obtained by template matching algorithm, and the algorithm is not robust, so the scope of application is limited

[0014] Fourth, the actual parking space detection environment for self-driving vehicles is complex and changeable, and the existing parking space detection methods are not robust enough to affect the automatic parking effect

[0015] Fifth, in the actual detection, it is inevitable to encounter occlusion and mis-matching of parking corners, resulting in false detection and missed detection, and the system is not robust

[0016] Sixth, when the actual collected images are 360° panorama stitching, the obtained parking space image will have a certain degree of "distortion", which will affect the accuracy of parking space detection

[0017] Therefore, the existing vision-based parking space detection methods are difficult to obtain satisfactory comprehensive effects in terms of scope of application, detection accuracy, robustness and real-time performance.

Images