TABLE PLANE DETECTION USING GEOMETRICAL CONSTRAINTS ON DEPTH IMAGES

Abstract

Plane detection is an important research in the robotics. Its results are utilized in many applications such as 3D reconstruction, scene analysis and segmentation, objects localization so on. In context of an assistive system for visually impaired people, plane detection, or more specialized, table plane detection is a prerequisite step for further procedures such as object localization and recognition. Although many approaches have been proposed for the plane detection such as RANSAC, RANSAC variants, Hough Transform, Least squares. Depending on the context of each application, selecting the appreciate results always required further implementations. But in fact, each specific application can only focus on one plane (interested plane). In this paper, we propose a new method for detecting an interested plane that is table plane in complex scenes based on depth images captured by a Kinect sensor. Our approach proposed algorithms combines PROSAC algorithm (an algorithm for estimating plane model) and geometrical constraints of environments. We compared some approaches together. The experimental results show that the proposed method outperforms the traditional ones.