Research On 3D Reconstruction Based On Binocular Vision

Stereoscopic 3D reconstruction has always been a hot topic in the field of computer vision.The paper builds a binocular stereo vision platform on the basis of reading a large number of domestic and foreign literature,and systematically studies the 3D reconstruction theory of discrete point based on binocular vision.The main contributions of this thesis are as follows:(1)Aiming at the core problem of stereoscopic vision-stereo matching,the paper compares several classic feature registration algorithms in current domain,analyzes the registration effect and running time,and The idea of using two kinds of characteristic detectors in combination with the complementarity of corner detector and patch detector is put forward.(2)On the basis of the previous research on different feature registration algorithms,a new algorithm combining improved FAST feature detection algorithm with average absolute difference template matching algorithm(MAD)is proposed,and the image acquired by binocular camera The experimental results show that the algorithm can achieve fast and accurate binocular image feature registration,which provides the premise for stereo matching to obtain parallax image in the next stage.(3)In the stereo matching algorithm,this paper systematically studies the stereo matching algorithm based on semi-global matching(SGM)and mutual information(MI),and combines the Oriented FAST-MAD feature extraction algorithm on a binocular vision platform to generate real-time disparity image and 3D point cloud data.Experiments show that the method has high efficiency,and the 3D point cloud information of the complete indoor scene is obtained,but the depth information of the object has some error and the error increases with the distance of the object.The main factors that cause error include system error,pixel quantization error and calibration error.For the above problems,the measured and actual values of the object at different distances are sampled.The least squares method is used to fit the sampling points,and the functional relationship between the actual value and the measured value is obtained.The function is used to correct the measured value,which effectively reduces the measurement error.Experimental results show that the depth information of the object after correction has a high accuracy.