In this paper we propose a method for segmenting single images from Kinect sensor by considering both color and depth information. The algorithm is based on a series of edge detection procedures designed for particular features of the scene objects. RGB and HSV color planes are separately analyzed in the first step with Canny edge detector, resulting in overall color edges mask. In depth images both clear boundaries and smooth depth changes are detected by using Canny detector and Sobel operator. Color and depth information are combined by merging detected edges into single contour mask, being the basis for initial segmentation. The final result is achieved by merging segments of similar depth. The method enables fairly precise scene segmentation with the ability of distinguishing objects of similar color or smooth boundaries.
Authors
Additional information
- Category
- Aktywność konferencyjna
- Type
- publikacja w wydawnictwie zbiorowym recenzowanym (także w materiałach konferencyjnych)
- Language
- angielski
- Publication year
- 2013
