Flexible fusion of 2D images and 3D time of flight data using plane segmentation

In the field of environment perception, image segmentation and region classification methods are of high interest. For the development of assistance systems based on the environmental situation, reliable information must be acquired. To achieve robust object detection, algorithms based on laser scanners or time of flight (tof) cameras outperform standard video image based methods. However, for region classification, appearance based approaches using video image data are best. Thus, a combination of both sensor systems can solve most problems of single sensor based systems. In this work, a combination of a tof camera together with a video sensor is presented to demonstrate the potential of complementary data fusion using both sensor systems. Using statistical knowledge, point clouds of the tof camera are segmented into 3D planes. Textures of flat objects are extracted for eased pattern segmentation. Resulting planes are classified with a monocular pedestrian detection algorithm and object dimensions are refined.