Binocular Stereo Vision Based Obstacle Detection And Alarm Methods In Reversing

With the spreading of the cars and the development of people's living standard,the cars bring us much convenience and lead to a lot of traffic accidents at the same time.Back is one of the most important parts of the vehicle driving,leads to accidents easily because of the blind area while backing.The existing reversing radar has the leak phenomenon,and reversing video can t calculate the distance between the cars and the obstacles.The monocular vision based obstacles' distance measurement will be impacted by the height and angle of the camera,So,the reverse auxiliary system based on binocular stereo vision is paid more and more attention.But,The binocular disparity image may easily effected by the regions whose tex ture are not that rich,may easily effected by the left regions of the left eye image,also,this is very time-consuming.The dispari,ty based obstacle divided is also very diffcult.All of the above could limit the widespread of binocular reversing system.The main content of this paper is binocular stereo vision based obstacle detection and alarm methods in reversing,in accordance with the following steps:a)binocular camera calibration,including the calibration of camera parameters which can be used for the calculation of obstacles;the calibration of the lowest obstacle's pixel height which can be used for the judgments of true obstacles;the calibration of the largest disparity for the binocular visual image which can be used for the cropping of left eye image and used for the stereo matching;the calibration of the distance lines which can be used as a reference of the obstacle's distance.b)Obtaining the disparity map,first we remove the road region whose texture are not that rich,and then get the exact cropping-valu e of the left eye image using the re verse-out ine--template matching method,then we use three types of methods to obtain the disparity map,obtaining disp arity map without the crop of left-eye,obtaining disparity map by cropping the largest-disparity-value of left eye image,obtaining disparity map by cropping the exact cropping-value of left eye image.In this paper,there are five kinds of stereo matching algorithms,the experimental results show that after cropping the exact cropping-value of left eye image,using the outline-template matching method to obtain the disparity map is the best choose.c)The splicing of the left eye image and the right eye image and the detection of suspected obstacles,image mosaic can take full advantage of binocular camera,and we use two methods to the detect the suspected obstacles:one is called the obstacles segmentation which is based on the disp arity map,including the multi-threshold segmentation of histogram and the UV disparity map method.Th,e second is called the obstacle detection based on the objects,the process is as follows,first we remove the road region whose texture are not that rich for the stitchingimage,and then detect the obstacles by get the outlines.The experimental results show that the obj ect-based obstacle detection method is the best;d)Obstacle distance calculation and the sound-light alarm system.Firstly,the pixel height of each suspected obstacle is calculated,then,with the calibration of binocular camera and the suspected obstacle' disparity value,we can calculate the pixels heigh t of the lowest obstacles which are corresponding to the suspected obstacles,we can judge the true obstacles by combining the pixel height of suspected obstacles and the pixel height of the lowest obstacle,With the calibration of binocular camera and the obstacles' disparity value,with the distance formula of binocular,we can calculate the distance of the obstacles,at las t,we marked the obstacles and do the alarming.Based on the above algorithms,an obstacle detection and alarm system in reversing based on binocular stereo vision were developed using Microsoft Visual C++6.0 and OpenCV library.The experiments show that these algorithms have a higher detection accuracy and a higher distance calculation precision.