Design And Implementation On Binocular Ranging System Combined With Strengthened Object

Stereovision is an important branch of computer vision research. In recent years, scholars pay more attention to stereovision for its broad prospect in various fields. This paper aims at designing a binocular ranging system combined with strengthened object based on OpenCV.Firstly, the brief overviews of OpenCV, coordinate systems, camera distortion, parallax and bokeh are presented in this paper which focuses on the basic methods of object strengthening and binocular ranging. To elaborate the design of binocular ranging system, the main parts are expounded as follows: 1) the checkerboard images are imported into stereo calibration which considers the homography of plane and makes use of OpenCV to calibrate the cameras; 2) after stereo rectification solves the epipolar lines such that the lines are in parallel and the two views are rectified, searching and matching corresponding points in two views are applied by the block matching algorithm; 3) analysis on the strengths and the weaknesses of Graph Cut, the image segmentation algorithm based on downsampling and morphological dilation is given in this paper, through which the system can select the object with simply user interaction. The size between any two points on the object can be computed and displayed on the bokeh image by mapping 2D points into 3D points.Then, the designed system is realized in the paper, which consists mainly of the contents listed as follows: 1) the calibration methods of OpenCV and Matlab are used to calibrate two cameras separately and the whole calibrating process is listed as well, the accuracy analysis of which are shown to estimate the quality of the checkerboard images and the experimental results demonstrate that only eligible images can lead to the correct calibration results; 2) stereo rectification uses the epipolar line constraint to compute the SAD of the rectified images. Data show that a smaller SAD leads to the excellent rectification results; 3) the given image segmentation algorithm is implemented in this paper. It is demonstrated from the results of different resolution that the speed of the given algorithm is three times at least more than the original algorithm; 4) with the projection matrix Q, the system can compute the coordinate of the real world point and display the size between any two points on the object in the final image.Finally, several images and analysis results show that the designed system can be used to range the object for the average error is less than 7% while the response time is less than 1 second.