Research On The Key Technology Of Binocular Stereo Vision Measurement System

Binocular stereo vision is a part of computer vision, researching how to mimic the human visual system with cameras and computer to realize the goal of perceiving the world around. Its applications involve driverless cars, biomedical, robot navigation and so on. In order to realize the function of extracting the depth information of the target object, the thesis designs an experimental platform of binocular stereo vision system. The system software programming is done in MATLABR2012 b environment.This thesis focuses on some key problems of the system of binocular stereo vision: camera calibration, image preprocessing, stereo matching and object extraction. First of all, the camera used in this thesis is calibrated, and calibration results are given; secondly, the images collected are preprocessed, including epipolar rectification, image denoising and edge detection; again, stereo matching is done after image preprocessing to obtain parallax map; finally, extract object on the parallax map, then calculate the depth of object according to the Similar-triangle-principle. The characteristics of this thesis are as follows:(1) Census transform of 8 neighborhood pixels is proposed based on the census transform of all, which is to extract 8 symmetric pixel in all the pixels within the neighborhood window as the pixels of census transform. At the same time, combine the census transform of 8 neighborhood pixels with region stereo matching and use it in the true scene. According to the theoretical analysis, for 13x13 neighborhood window, the times of comparison of the improved algorithm are one over twenty-one of the times required for the original algorithm. By experimental verification, the speed of the improved algorithm is 3.3 times faster.(2) An algorithm of the extraction of target object is proposed, which is to get the parallax map containing target object only by using binarization processing to original parallax map, connecting region labeling and removing small areas on the parallax map, then calculate the depth of object according to the Similar-triangle-principle. By experimental verification, this method can avoid the complexity of 3D reconstruction and obtain the distance of object directly. So it has some practical value.Experiments are done in the experimental platform to test the system and the experimental results show that the system can do well in image acquisition and also be stable for image processing and getting the depth information of the target object...