Adaptive visual odometry using RGB-D cameras

Abstract

An adaptive color-depth (RGB-D) visual odometry algorithm is presented to enable high-accuracy egomotion estimates while reducing computational performance. Specifically, the presented algorithm uses a statistical confidence interval to adaptively ensure accuracy of the visual odometry solution while at the same time controlling the computational performance. This in turn reduces the computational requirements of implementing the algorithm. Experimental studies presented in this paper show that this adaptive algorithm can achieve an error of 0.8% with reduced computational load.