STEREO-BASED RECONSTRUCTION UNCERTAINTY AND EGO-MOTION ESTIMATION

Abstract

Based on the analysis of reconstruction uncertainty, an ego-motion estimation method using the uncertainty to reduce error propagation is proposed. In our method, Harris corners are matched and tracked by an improved MNCC algorithm which is proposed and utilized to reduce illumination effect. The matched and tracked corners are reconstructed by least-squares method, and the reconstruction uncertainty is described by covariance matrices. least-squares method and sparse Levenberg-Marquardt method combining the reconstruction uncertainty are utilized to estimate the motion of mobile robot roughly and more accurately, respectively. Simulation results showed that our method can estimate the motion of mobile robot accurately and reduce the error propagation efficiently. Practical experiment results in both outdoor and indoor environments demonstrate the actual validity of our method.