Support-plane estimation for floor detection to understand regions' spatial organization

Plane fitting plays an important role in image processing and computer vision. It is challenging because of the outliers that do not follow the plane pattern. In this work, we address the problem of support-plane fitting for room floor detection from point clouds that are generated from depth image. Based on the geometric layout of data, an optimization problem is derived to estimate the support-plane. Algorithms are also proposed to deal with data noise. The floor detection is achieved by support-plane fitting, and is employed as a reference to analyze the spatial organization of room scene. A projection method is presented to form the organization map. Experiments demonstrate the proposed method is more robust, and it achieves remarkable performance in understanding the spatial organization.