Continuous Plane Detection in Point-cloud Data Based on 3D Hough Transform

In modern robotics vision systems, a shape extraction from point clouds (or depth images) is an important and still discussed  topic. We present significant optimizations of 3D Hough Transform for a plane extraction from point cloud data. The method aims to overcome noise present in these point clouds, high memory requirements for the parameter space and computational complexity of point accumulations. All these problems are further discussed and solutions are proposed to design a robust plane detector. The detector benefits from the key principle of the Hough transform, the accumulation of values in the parameter space, and processes continuous point cloud stream from the depth sensor for iterative refining of results. The proposed technique is compared against the optimized implementation of RANSAC-based plane detector in the well-known PCL library.