Research on Visual Odometry of Planar Motion Wheeled Robot Based on RGB-D Camera

Based on the motion characteristics of the planar motion wheeled robot with RGB-D camera, the contour constraint conditions were proposed in the image processing part of visual odometer to screen out the mismatched feature points, and a new calculation method was proposed in the inter-frame motion calculation part. Firstly, the contour constraint condition was established according to the motion constraint of the planar wheeled robot camera. This constraint condition could not only screen out the color map mismatch, but also screen out the matching point pairs with correct color map matching but large depth value error. Then, in the inter-frame motion calculation part of the camera, the transfer matrix was reduced in dimension and the least squares problem was constructed on the basis of linearization approximation. Finally, the linearization approximation calculation error was dynamically compensated to obtain the two-dimensional motion parameters of the robot camera. The experimental results showed that compared with RANSAC algorithm, the algorithm exhibited higher calculation accuracy. Compared with RANSAC algorithm and ICP algorithm, the operation time of this algorithm was reduced by
$89.08{{\% }}$
and
$52.38{{\% }}$
, respectively.