Research And Implementation Of Parking Assist Methods Based On Computer Vision

With the development of social economy,cars have become daily consumer goods.Parking assist system is a concrete manifestation of intelligent vehicle,drivers can avoid a collision and achieve safety parking through the parking assist system.Especially,the parking assist system based on computer vision has good vision and high reliability,it has become the forefront of current research.The parking assist system based on computer vision extracts the parking contour and parking entrance by using image processing technology,according to the parking entrance coordinate,the system implements ranging function and path planning to provide a safe and reliable parking path for driver,it effectively reduces parking economic losses and casualties.The main work of my thesis is as follows.(1)The thesis processes the parking image,including image gray,image denoising and image dilation and erosion.Image gray changes a color image into a gray image in order to reduce the computational complexity;F-BMDA combine bilateral filtering and median filtering in order to removes complex noise;Image dilation and erosion ensure the parking contour integrity.(2)The thesis extracts the parking contour by using otsu algorithm.Firstly,the thesis analyzes the feature of parking contour;then,uses iterative method and otsu to extract the parking contour,experimental results show that otsu algorithm can extracts parking contour effectively.(3)The thesis extracts the parking entrance coordinates according to the corner feature of parking contour.Firstly,the thesis analyzes the corner feature of parking contour,then proposes a feature template to extracts the entrance coordinates.In order to improving the accuracy of extracting parking entrance,the thesis improve the algorithm according to the horizontal parking contour projection and vertical parking projection.(4)The thesis proposes a Mean-Shift tracking algorithm based on contour corner feature in order to effectively solve the shortcomings of traditional Mean-Shift that loses the target due to the change of the target's size,experimental results shows that the improved algorithm can track parking entrance accurately.(5)The thesis analyzes the ranging monocular model and implements model in the parking entrance distance measurement's applications.According to car's maximum deflection angel,the thesis achieves parking planning of theoretical level parking and vertical parking analysis combined with distance information.