Feature extraction techniques for lidar range profile based object recognition

The article provides an overview of range profile feature extraction methods used in laser identification, detection and ranging systems. It also outlines feature selection methods and highlights their respective limitations. A novel feature selection method which maximizes Euclidian distances between feature vectors is presented. The article also showcases advantages of the proposed technique by extracting features of basic objects (a sphere, a cone, and a cylinder). This method is shown to be effective when feature vector manifolds are not linearly separable due to the unknown viewing aspect of an object. The technique is also effective when feature vector manifolds overlap due to the different objects having similar range profiles.