Research And Implementation Of Binocular Vision Stereo Matching Algorithm

Computer vision is a research hotspot in the field of artificial intelligence.As an important branch of computer vision,binocular stereo vision enables computers to perceive the real world by simulating the human visual system.It is an important means to obtain depth information of real scenes.It plays an important role in visual navigation,3D reconstruction,augmented reality,virtual reality,automatic driving,and so on.Stereo matching technology uses the method of pixel matching to estimate the disparity of image,obtains disparity map according to matching relationship,and finally obtains the depth information of real scene.It is the most popular field in stereo vision research.The performance of stereo matching algorithm will directly affect the quality of disparity map and the effect of final depth perception and 3D reconstruction.Therefore,the stereo matching algorithm with high precision,fast speed and strong robustness is one of the research hotspots.The main research contents of this paper are binocular stereo matching algorithm and its application in depth perception and 3D reconstruction.Two stereo matching algorithms are proposed from different research perspectives,and the application platform of binocular stereo vision is built.The main innovations are as follows:(1)Aiming at the problem that traditional stereo matching algorithms are sensitive to noise,easily distorted and have high false match rate in disparity discontinuous regions and weak texture regions,a multi-scale stereo matching algorithm based on improved Census transform and gradient fusion is proposed.The weighted average gray value of all pixels in the support window is used as the reference value of Census transform,and the Census cost is weighted and combined with the gradient cost normalized by the horizontal and vertical directions.Under the multi-resolution scale,the improved guided filtering algorithm is used to complete the matching cost.The experimental results show that the proposed algorithm is robust to noise and illumination disturbances,and the mismatch rate between disparity discontinuous region and weak texture region is further reduced.(2)Aiming at the problem that the existing local stereo matching algorithms can not distinguish strong and weak texture regions well and can not solve disparity ambiguity in disparity calculation,a stereo matching algorithm combining gradient feature and confidence is proposed.The support window of matching cost calculation is selected according to gradient feature information,and different sizes of support windows can be selected for strong and weak texture.Confidence is introduced in the disparity calculation,and disparity singular value is corrected by horizontal and vertical cross region detection.The experimental results show that the proposed algorithm is suitable for the cost calculation of different texture regions,effectively resolves the disparity ambiguity problemin disparity calculation,and improves the robustness and accuracy of stereo matching algorithm.(3)The application system of depth perception and 3D reconstruction based on binocular stereo vision is designed,and the application platform of binocular stereo vision is built.The method introduced in this paper is used for stereo calibration and correction.The disparity map in real scene is generated by the stereo matching algorithm proposed in the paper,then the depth perception and 3D reconstruction of real scene are realized according to the disparity map.For the application of the real scene,the perfect experimental scheme design and sufficient experimental verification and analysis are carried out.The experimental results show that the binocular stereo vision system has high depth perception accuracy and good three-dimensional reconstruction effect,which fully verifies the accuracy of depth perception and the feasibility of three-dimensional reconstruction,and also shows the feasibility and practicability of the research content in this paper.