Developing a Novel Stereovision Scheme for Improving Accuracy of Depth Estimation in Robotic Vision Systems

Abstract

Acquiring three-dimensional data from a pair of stereo images is called stereovision that has been studied by researchers for decades. However, most of the previous studies on this topic focused on establishment of stereovision parameter matching and made conclusions on the premise of fixed focus. With the rapid development of multimedia technology, varifocal digital cameras have been widely used in many robotic applications recently. In general, error in the depth estimate becomes bigger when the focus and aperture is unknown or not fixed. For that reason, a three-stage framework is proposed in this paper to modify the conventional stereovision model for improving accuracy of depth estimation. The first stage is to modify the computational model of conventional stereovision for varifocal cameras. Then, the spacing of depth interval in non-uniform spacing of discrete depth levels can be altered, in particular, it is unaffected by changes in focal length. Finally, with considering the affine transformation, we add the deformation coefficient into the modified stereovision model for correcting three-dimensional affine deformations. Experimental results demonstrated that the depth estimation from stereo images using the proposed scheme was more accurate than conventional method.