Research On Obstacle Detection Based On Binocular Vision

With the progress of society, the growing populations made the human beings face a lot of agricultural problems. In order to solve these problems, agricultural production of intelligence has been widely studied, especially agriculture autonomous vehicle. There are a lot of ineviTab.le obstacles appearing in actual farmland environment, so how to detect the barrier is a very important problem. In all kinds of sensors, with a moderate price, binocular camera can easily get the depth information of objects in the scene and be able to get enough environmental information, and ordinary computer can adapt its processing speed. Furthermore, binocular camera is similar to human eyes’ structure, so it conforms to cognitive habits of people and is more conducive to the use and study of people. Therefore, based on the binocular camera, this paper studied the obstacle detection based on binocular vision.This paper firstly introduced the background of this research and its significance, elaborated the present situation of research on obstacle detection technology at home and abroad and put forward the main research contents of this paper; Then, from the overall design of binocular vision, the paper put forward the whole structure and implementation steps of binocular vision system, and from two aspects of hardware platform and software platform, the binocular vision system was constructed.In the aspect of camera calibration, this paper introduced the principle of camera imaging and several reference coordinate systems required for camera calibration, it also made conversion between the relationships of coordinate systems. For camera calibration method, this paper mainly analyzed three traditional calibration methods:DLT calibration method, Tsai calibration method, Zhang Zhengyou plane calibration method. Because Zhang Zhengyou plane calibration method has the advantages of simple operation, real-time and high robustness, thus this paper used Zhang Zhengyou plane calibration method and got the camera internal and external parameters through camera calibration experiments by MATLAB software.In the aspect of extraction of the obstacle target area, this paper mainly adopted the maximum between class variance method to extract the optimal threshold, made image segmentation, and got the approximate area of the obstacle.For feature point detection and matching of binocular camera images, this paper mainly introduced three kinds of typical algorithms:Harris-NCC algorithm, SURF algorithm, SIFT algorithm. And by using the three kinds of algorithm for image feature point detection and matching, it analyzed and compared the performance of the three algorithms. These performances include the number of detection of feature points, detection time for feature points, the number of corresponding points successfully matched and matching time. Through analysis and comparison, it concluded that SURF algorithm is the best algorithm for feature point detection and matching.Finally, this paper calculated the depth information. It mainly selected feature points calibrated by obstacle objects through the corresponding points matched by SURF algorithm. Then it calculated the depth information of obstacle target according to the relationship between the world coordinate system and the image coordinate system and the internal and external parameters calibrated. Through the experimental analysis, there are some errors between calculated depth and distance measured, and error rate becomes smaller gradually as far away from the direction of the robot car.