Research On The Parking-slot Detection Algorithm Based On The Around View Monitor System

"Autonomous driving",the mother of artificial intelligence problems,has been at the forefront of the wave of artificial intelligence.According to the survey,in all intelligent driving assistance systems(ADAS),automatic parking becomes the most urgently needed intelligent driving assistance system for users.However,most of the production vehicles equipped with automatic parking systems use ultrasonic as the parking space detection method.Since the ultrasonic detection parking space can only recognize the parking spaces with vehicles at both ends and cannot recognize the parking spaces without vehicles at both ends,it has great limitations in the application scenario.In addition,the existing parking space detection algorithms mostly use the traditional digital image processing methods,and the algorithm has low precision and poor robustness,so it has not been applied to the actual scene so far.With the development of image processing technology,the around view monitor system has gradually become the standard for mid-to high-end models,and the application of the parking space recognition methods based on the around view monitor system to the automatic parking system has gradually become the future development direction,thus improving the accuracy of parking space recognition based on around view monitor system is of great significance for perfecting the automatic parking system.Firstly,this paper uses Prescan software for scene construction and algorithm verification.Four large wide-angle fisheye cameras are set up by simulation software,and a variety of complex scenes are built to verify the effect of the loop-splicing algorithm and the parking space recognition algorithm,and the collected data will be used as training samples extracted from the deep learning region of interest.Secondly,the problem of complex image registration process when splicing the car’s around view monitor system is improved.For the problem that the artificial locating point is easy to produce precision error in the algorithm of the double longitude model algorithm,the Harris feature detection method is added to automatically extract the feature points and use OpenCV performs calibration to solve the internal and external parameters of the camera.Then,a new method of extracting the region of interest is proposed.The semantic segmentation method in deep learning is used to dynamically extract the parking space part of the whole image.The method simultaneously outputs the rectangular frame and the pixel level mark.The rectangular frame is used to extract the parking space parallel to the vehicle body,and the pixel-level marking is used to extract the parking space with a certain inclination angle to the vehicle body.Finally,using the digital image processing method,the contrast between the parking space and the background is improved by the contrast enhancement method.Then,the edge detection,Hough transform and clustering method are used to extract the four corner points of the last parking space to realize the function of parking space recognition.