The algorithm's first step is to estimate the orientations of individual points in the scene (orange arrows), which it maps onto the surface of a sphere (orange clusters). Through an iterative process, it finds the set of axes that best fit the point clusters (red, blue, and green columns), which it re-identifies with the points in the scene. Image courtesy of the researchers
Read: Orienteering for Robots
The researchers' new algorithm takes depth information (red) about a visual scene and determines the orientation of the objects depicted (red, blue, and green). Courtesy of the researchers
That makes the problem of plane segmentation — deciding which elements of the scene lie in which planes, at what depth — much simpler (multiple colors). Courtesy of the researchers
For more information visit http://web.mit.edu.