A Quadrant Approach of Camera Calibration Method for Depth Estimation Using a Stereo Vision System

Stereo vision systems are well know depth estimation methods with a large number of applications such as automatic inspection, autonomous navigation, process control, etc. The functioning principle of these systems is the triangulation between the real-world surface point and its respective projections on the image planes of each camera. One of the key points in order to obtain accurate measurements on stereo vision systems are the calibration of extrinsic and intrinsic parameters. This is why the work of this paper focuses on a camera calibration method to correct the error generated by the lens distortion. The proposed method divides the image in quadrants and generates an equation for each quadrant to correct the error generated by the lens distortion. The performed experiment demonstrated an accuracy improvement using the calibration method compared to the measures taken without a calibration method.

I. Introduction

Stereo vision is an active research topic with many applications among different fields like in-hand object localization [1], 3D imaging of underwater scenes [2], autonomous navigation [3], automatic pose estimation [4] etc. These applications of stereo vision systems require precise measurements, and in order to obtain them, several factors must be taken into account like stereo disparity, image planes geometry, camera calibration [5], etc. Stereo disparity is the task of finding the correspondence between an object or a pattern in both stereo images, some methods used to perform this task include SAD [6], SSD [7], NCC [8] and deep learning techniques [9]. The image planes geometry [10] refers to the intrinsic [11] and extrinsic [12] parameters of the cameras such as sensor size, distance between cameras, angle of cameras, focal distance, etc. The camera calibration is the task of correcting the image distortion generated by the camera lenses [13].