A Fast and Deterministic Plane Detection Method in Point Clouds based on Hough Transform

抄録

3D Plane Detection in Point Clouds has many important applications, such as Object Recognition and Robot Navigation. The state-of-the-art approaches are namely RANSAC and Randomized Hough Transform, although, due to their non deterministic nature, the finishing conditions have to be chosen carefully, hence, these approaches tend to not have a defined processing time . Therefore, we propose a method that speeds up the plane detection process in a deterministic way by looking at the centroids of individual patches of the Point Cloud. Then, a refinement of the decision is made by looking at the eigenvalues of each patch. Consequently, Hough voting is performed to the plane calculated from the coplanar patch. The experiment results have demonstrated that this method outperforms state-of-the-art approaches while being deterministic.