Topological structures learning using the Delaunay triangulation and cascade support vector machines Mohamed cherif Rahal May 27, 2014

Institut VeDeCoM, 77 rue des Chantiers, 78000 Versailles [email protected]

Abstract Reliable object recognition is an important step for enabling mobile systems to reason and act in the real world. A high-level perception model ables to integrate multiple sensors can signicantly increase the capabilities of such systems. Tasks such as obstacle avoidance, mapping, and tracking can provide real benets from a fast and general objects detector able to recognise and identify relevant objects from training stages. Our problematic deals with the recognition of objects on the way of sensors-equipped mobile system. Most of the existing vehicular and robotic sensors provide a set of points in R3 at each time period. Having this huge amount of data organised as timestamped sets of points, our aim is to cluster, learn and recognise structures from each set of points. However, these clustering and recognition tasks must be done in real time and without any notion about the number of desired clusters. The problem is clearly related to srstly an unsupervised task folowed by a supervised learning task. In the clustering stage, rst we compute the delaunay triangulation. Then we apply the α-shape algorithm using a suitble parameter α. Finally we extract the clusters by using connected component graph-based research approach.For each connected component which represents an object in the scene, we extract a certain number of features. After the clustering and features extraction phase, a labeling and learning tasks are needed to create the learning model. For the last stage we use a cascade SVM based technique.

Fig.1 An Example of the application of the α-shpa algorithme on a 3-d point cloud provided by a velodyne laser-scanner of 64 layers.

References [1] Dominique Gruyer, Aurélien Cord, Rachid Belaroussi (2014): Vehicle detection and tracking by collaborative fusion between laser scanner and camera. IROS 2013: 5207-5214. [2] Edelsbrunner, H., Mucke, E. P. Three-Dimensional Alpha Shapes. ACM Transactions on Graphics 1994 , 13(1), pp.43-72. [3] Chen, Z., Pears, N., Freeman, M., Austin,J. : Road Vehicle Classication using Support Vector Machines. ICIS 2009. IEEE International Conference on Intelligent System(Volume:4). 1