indoor segmentation and support inference from rgbd images

aim:

In this paper, our goal is to provide such a physical scene parse: to segment visible regions into surfaces and objects and to infer their support relations. In particular, we are interested in indoor scenes that reflect typical living conditions.

difficulties:

prevalence of small objects, and heavy occlusion, which are all compounded by the mess and disorder that are common in lived-in rooms.

advantages:

large planar surfaces, such as floor, walls, and table tops, and objects can often be interpreted in relation to those surfaces. estimating the floor orientation or finding large planar surfaces are much easier with depth information.

key technology: