Multi-camera stereo vision based on weights

Abstract

The improvement of measurement accuracy has always been a hot topic in visual measurement. The multi-camera stereo vision, which is composed of more than two cameras, provides more image information, stronger interference capability and higher 3D reconstruction accuracy than binocular vision, has been widely used. The imaging quality, camera calibration accuracy and vision system structure parameters of different cameras may be different. However, in the traditional multicamera stereo vision, the contribution of each camera to the reconstruction results is the same, which may lead the reduction of the reconstruction accuracy. In this paper, multi-camera stereo vision based on weights is proposed to reduce the impact of cameras with large errors, eventually, the measurement accuracy is improved. The error characteristics are analyzed comprehensively, and the error model is established to calculate weights, then the weighted least square method is used for 3D reconstruction. The feasibility of the proposed method is verified by the trinocular vision 3D reconstruction experiment. Compared with the traditional 3D reconstruction method based on least square method, the accuracy is improved by about 3%.