Error Modeling of Depth Measurement Using FIR Stereo Camera Systems

Abstract

This paper deals with an error model and its evaluation in estimating the 3-D depth information using stereo camera system. This paper defines an error model caused by stereo matching, and evaluates the error bound empirically in estimating the depth information. The proposed error model is that the disparity values in the pixel domain have ±0.5 pixel errors maximally even though the disparity values are correctly estimated. Thus, the depth information also has the same amount of errors with respect to the focal length and baseline of stereo cameras. To evaluate and apply the proposed model to the military applications, we use uncooled FIR cameras to set the stereo camera system and consider the small high-temperature objects as the targets. We first detect the high-temperature moving object in the stereoscopic IR images, and estimate the disparity field in the pixel domain. Then, we calculate the physical depth between the cameras and moving object by using triangulation of disparity, baseline and focal length. According to the experiments with various distances and baselines, we evaluate the error bound in the depth estimation. This error bound and stochastic parameters are useful for estimating the 3-D position of moving objects probabilistically.