Research On Monocular Vision Automatic Parking Control System

According to statistics, up to the end of2012, the number of cars in China has reached120,000,000, and the number is expected to continue to grow. The substantial increase in thenumber of cars to make the already congestion parking spaces become more narrowed.Parking smoothly in the narrowed spaces is difficult to many drivers, resulted in the increaseof the accidents year by year. Machine vision parking assist system can effectively solve theseproblems. However, researched machine vision parking assist system has the drawback ofrequiring manual intervention and poor slot marking recognition. This paper proposes thealgorithm of monocular vision automatic parking control system aiming at these problems.The main contents of this paper are as follows:(1)compared advantages and disadvantages of image measurement using monocularvision and binocular vision. Calibrate camera with Zhang Zhengyou camera calibrationmethod. Analyzed the effects of camera position changes to measurement results.(2)Build test platform for monocular vision automatic parking control system. Controlthe movement of car using Freescae’s MC9S12XS128chip as car control MCU. Process avideo signal alone with TI’s TMS320DM642chip as a coprocessor,. Designed the sensormodule and actuators, together constitute the hardware structure of the automatic parkingcontrol system, and designed the automatic parking control system software flow.(3)Studied deeply monocular vision-based slot marking recognition algorithm consistedof two parts, namely rough extraction of spaces corner and precise matching of spaces corner.Firstly, preprocessing the image, and converts it into a bird’s-eye view. The idea of thepyramid hierarchical was introduced, and slot marking skeleton in the top pyramid image wasextracted. Detected skeleton using Hough transform, and determined parking corner candidate.Then, according to the location of parking corner candidate cut the region of interest in thebird’s-eye view, and extracted skeleton of parking corner in every region. Adaptive geneticalgorithm is used for skeleton template matching. Obtained the image coordinates of allparking corners and determined the target parking spaces based on the geometriccharacteristics of the real spaces. Lastly, test of slot marking recognition with50sampleparking spaces under different lighting conditions had been done and an accurate rate of98%was achieved. The experiment results showed that proposed slot marking recognitionalgorithm could effectively identify the target parking space.(4)Some preliminary exploration had been done about path planning. Firstly, introduced briefly arc-based parallel parking path planning. Then, vehicle kinematic model based onAckermann steering geometry was established. Parallel parking path with continuouscurvature based on clothoid was planned according to car’s steering characteristics andparking information, which avoided the shortcomings of the curvature discontinuity and hadan important significance for improving parking efficiency and increasing operator comfort.(5)Functional effectiveness of test platform was verified conducting parallel parking testusing model car, which improved and developed conveniently the follow-up controlalgorithm.