Alignment and Super Pixel Segmentation of RGB-D Video Stream
Lianjun Liao, Yongbin Hao, Xiangyang Su, Shihong Xia · 2016
In this work, the RGB-D video stream is captured from Microsoft's V2 Kinect, and is used for alignment and super pixel segmentation. We found an effective method to align the RGB-D video stream. Then the aligned depth video stream is optimized by the joint-bilateral filtering algorithm. The 3D scene can be reconstructed by the time and space alignment of the RGB-D video stream. Moreover, we proposed a new segmentation method for RGB-D video stream, which uses the K-means clustering method to produce super pixels. For the first time, we introduce the optical flow information of the video stream into super pixel segmentation. With the position, color, depth, and the optical flow information between the front and back frames of the color video, our new algorithm can make a more accurate super pixel segmentation. Finally, we do several experiment to demonstrate the effect of augment method.