Edge and Intensity based Visual Odometry for RGB-D Camera

We propose a novel visual odometry based on edge and intensity for RGB-D cameras. The proposed method is based on frame-to-keyframe tracking, which aligns the edges extracted by the Canny detector and minimizes photometric errors to estimate the camera motion. Edge alignment is realized by minimization of distance transform (DT) errors which are calculated by the edges. The huber robust cost function and the Student-t distribution are applied to weight the DT errors and the photometric errors respectively. A joint optimization for DT errors and photometric errors is generated for pose estimation by nonlinear optimization. The method is evaluated with the TUM RGB-D dataset. Our method can run in real time on a laptop computer.