Analysis of Ranging Error of Parallel Binocular Vision System

The problem of ranging error has posed great challenges for 3D reconstruction and obstacle avoidance in parallel binocular vision system. For robot platform, parallel binocular vision system calibrated in fixed distance leads to low accuracy in depth estimation when distance changes. The error transfer function model in the system is applied to solve the problem. However, the theoretical ranging error along central axis of binocular cameras decreases with increasing depth and eventually converges to a limit value, and the ranging error near the central axis is sensitive to a minor change of field of view (FOV). These results are inconsistent with real scene. In this paper, an innovative error analysis method is proposed by introducing spatial uncertainty due to discretization of pixels to geometric model. The proposed method indicates theoretically that the ranging error of object within common FOV along the central axis grows with increasing depth and the magnitude of ranging error is modified. Also, experiments are implemented to further validate the reasonability of the proposed method.