Evaluation of the quantization error in convergence stereo cameras

Abstract

Computer vision techniques including image acquisition, image processing, and data analysis are currently popular and widespread. Stereo vision systems in the area of computer vision have been widely used in many applications, such as 3D depth reconstruction, 3D gaming, robot vision navigation, remote sensing, medical image processing, and so on. However, highly accurate 3D stereo vision scanners do not completely satisfy industry requirements due to the high price of these systems. In this paper, what we believe to be a novel method is proposed that increases the 3D measurement accuracy of stereo vision systems with low resolution and cheap cameras. This was achieved through discussing the quantization error of the 3D depth reconstruction of stereo cameras. Accordingly, a mathematical model was introduced to evaluate the quantization error of convergence stereo cameras. Meanwhile, a synthetic experiment was employed to validate our proposed mathematical model. Finally, simulation experiments were conducted to demonstrate how to apply the presented approach to increase the accuracy measurement of low resolution stereo vision cameras. The simulation results show a considerable measurement improvement by changing the convergence angle of stereo vision cameras.