Research On Automatic Parking Technology Based On Machine Vision

Automobile has already been an essential tool in people’s daily lives. Along with the development of automobile industry and computer technology, automotive advanced driver assistance systems, or ADAS, which is significant in reducing traffic accident, improving city transport efficiency, is gradually being intelligent. As a typical kind of ADAS, automatic parking system will be widely applied in automobile in near future. Automatic parking system generally comprises of multi-ultrasound or multi-radar arrays, which are both very expensive. Automatic parking system is composed of sensing module, decision module and control module. The sensing module is the input of the whole system and the key to complete parking successfully. Parking space can be recognized accurately and efficiently from complex background in color image by the algorithm which designed in this dissertation. It is significant in popularization of automatic parking technology.Firstly, parking space line is difficult to be extracted from complex background. Algorithm for extracting yellow parking line from complex background image has been designed. Color image contains rich information compared with gray image. Color feature of parking space lines and background are very different. Compared with sensing module based on multi-radar arrays, computer vision sensing module can detect not only vacant parking space between two automobiles but one in parking space lines. The experimental results show that parking lines can be segmented from background successfully by algorithm mentioned above.Secondly, parking lines may be dirty and lose original color feature in some parts. The algorithm based on mathematical morphology has been designed to mark the vacant parking area inside lines. Parking line on binary image may be discontinuous and these gaps can be bridged. Then the target area can be extracted from complex background by this method. The experimental result shows that the target area can be marked and the four coordinates of the vertices of the area can be extracted accurately.Finally, verification and test of parking area recognition algorithm on real automobile is very difficult. 1:12 scale experiment platform and testing environment has been designed and implemented. The experiment platform consists of Android smartphone and 1:12 scale model car. Android smartphone, which can be installed on the model car, can capture and processing image. And it is also able to communicate with model car by WiFi LAN. The design is low-cost and easy to test automatic parking algorithm. The Android smartphone installed app of parking area recognition can also work alone and test algorithm in real parking lot.Eight random parking slots has been chosen to test the algorithm in both A and B parking lot. Algorithm program has been tested 30 times on each parking space. Vacant parking area can be extracted and the average success rate is more than 90%.