An approach to loop-closing based on RGB-D images

To solve the problem of accuracy and robustness seriously affected by the external environment when loop-closing using vision for mobile robot, a method based on RGB-D image for loop-closing detection is proposed. The front area contour which will be used for contour matching is extracted from the depth information. Then Image matching between similar scenes of candidates is done Using PCA-SIFT descriptors, and in which the same scene is collected to achieve loop-closing. This method is suitable for the loop-closing in the poor lighting or even dark complex indoor environment, and effectively improves the detection efficiency and accuracy.