Range image based classification system using support vector machines

Abstract

This paper describes a classification system based on Support Vector Machines (SVM) and using 3D range images. Two kinds of camera systems are used to provide the classification system with 3D range images: Time-oF-Flight (TOF) camera and Stereo Vision System. While the former uses the modulated infrared lighting source to provide the range information in each pixel of a Photonic Mixer Device(PMD) sensor, the latter employs the disparity map from stereo images to calculate three dimensional data. The proposed detection and classification system is used to classify different 3D moving objects in a dynamic environment under varying lighting conditions. The images of each camera are first preprocessed and then two different approaches are applied to extract their features. The first approach is a Computer Generated method which uses the Principal Component Analysis (PCA) to get the most relevant projection of the data over the eigenvectors and the second approach is a Human Generated method which extracts the features based on some heuristic techniques. Two training data sets are derived from each image set based on heuristic and PCA features to train a multi class SVM classifier. The experimental results show that the proposed classifier based on range data from TOF camera is superior to that from the stereo system. © Springer-Verlag Berlin Heidelberg 2007.