Pedestrian Recognition Based on 3D Image Data
Bjorn Elias, Petri Mähönen · 2007
This paper presents a pedestrian classification system using 3D imaging information. The system consists of a photonic mixer device sensor (PMD), infrared lightsources and image processing algorithms. We present an example of an active safety system to protect pedestrians, describe a pedestrian classification using 3D image image processing, and give suggestions for further optimization of the software. The focus of this work is the description of main features which are required in order to accomplish a reliable classification. The task of the classification algorithms is to classify sensor objects either into the category pedestrians or the category non-pedestrians. After taking a closer look at image processing and measurement results we draw the conclusion that a multisensor approach is an inevitable requirement in order to fulfil the strict requirements of an exemplary active safety system given in this paper. Finally, the paper explains improvements of the classification on the basis of feature level fusion using 2D and 3D sensor systems.