Dynamic Object Detection Using Improved Vibe for RGB-D SLAM

Simultaneous localization and mapping (SLAM) is essential for autonomous navigation of mobile robot. However, dynamic objects, such as walking people, can seriously degrade the performance of SLAM. An improved Vibe (IVibe) algorithm that bases on background frame updating is proposed to detect dynamic objects in RGB-D SLAM. The background frame is updated according to the occupancy rate of foreground points and the changing value of the depth pixels. The homography matrix is used to eliminate the space mismatch between the background frame and the current frame. The foreground point inspection method is applied in pixel classification to deal with the "ghost" point. The experiment results show that the IVibe-based approach can detect the dynamic objects in RGB-D SLAM effectively.