Visual odometry based on the pixels grayscale partial differential

The visual odometry based on Photo-Consistency assumption is affected by noise and texture. The pixels grayscale partial differential is applied to overcome noise effects and enhance image texture in the field of image processing. This paper proposes a visual odometry method by calculating pixels grayscale partial differential. The algorithm makes full use of the continuity of image information and utilizes the information around a pixel to register images, improves performance of the visual odometry. Experimental results on the TUM RGB-D datasets show that the presented method can achieve better performance on multiple types of image sets, especially in texture-less scenes.