Vision Processing Technology Based On Binocular Camera

With the advancement of the era of digitization,informatization and intelligence,the research of binocular stereo vision technology is of great value.Benefiting from its advantages of simple system structure,easy implementation and low cost,binocular stereo vision technology is widely used in intelligent transportation,unmanned driving,navigation and other scenarios.However,due to the poor matching accuracy of binocular stereo vision system and its difficulty in recovering depth information of complex scenes in the practical application scenarios,how to improve its matching accuracy has become the focus of current research.In response to the above problems,based on the system platform built by Bumblebee2 binocular camera this paper mainly conducted in-depth research on the core technologies of Bumblebee2 binocular camera calibration,image defogging preprocessing,stereo matching,etc.The main research contents were as follows:(1)First of all,through the systematic study on the camera imaging model and binocular vision theory,the hardware platform for binocular vision was built with the Bumblebee2 camera,and the software experiment environment of Fly Capture and Triclops SDK was configured.The common calibration algorithms of cameras were compared and analyzed,and the binocular camera calibration and stereo calibration was carried out by MATLAB calibration toolbox based on Zhang's calibration method.Finally,we obtained the calibration parameters through experimental simulation,and achieved the expected effect.(2)In order to solve the difficulties in visual processing of outdoor scenes such as unmanned driving and intelligent transportation in severe foggy weather,an in-depth research was carried out on image defogging technology in the image preprocessing stage from the following aspects: the first is to deeply study and reproduce the dark channel prior algorithm in the image defogging algorithm;the second is to propose an improved dark channel prior dehazing algorithm aiming at the problem of halo and distortion of the image in the sky area after the original image was dehazed by the dark channel prior algorithm;the third is to segment the original image containing the sky area employing Otsu algorithm,reasonably estimate the atmospheric light value through the segmented sky area,and further optimize the transmittance of the non-sky area by gradient guided filtering.The experimental results show that the improved defogging algorithm can well restore a clear image,and the large deviation of atmospheric light value estimation and the color distortion in the sky area are effectively improved,which lays a foundation for the subsequent visual processing.(3)After an in-depth study on the local stereo matching algorithm,a stereo matching algorithm based on the improved Census transform method was proposed for the problem of poor matching accuracy of the local stereo matching algorithm in weak texture and deep discontinuity regions.First,the improved Sobel operator was adopted to calculate the edge information of the image,which would be used as a reference value to adaptively select the window for census transformation.Secondly,the matching cost was calculated by fusing the image gradient information,and the matching cost was aggregated with the minimum spanning tree method.Experimental results show that the algorithm in our study obtains higher image matching accuracy and better matching effect than other local matching algorithms.Finally,the algorithm of this study was applied to the binocular vision system to obtain the binocular images in the real scene,and complete the three-dimensional reconstruction of the real scene in line with the principle of three-dimensional reconstruction.