Interpretation Of Traffic Scenes Using A Hierarchical Data Structure

Abstract

Within the goals of PROMETHEUS program', safety plays an important role. Observing and interpreting the environment continuously promises a significant reduction of accidents caused by lack of alertness of the driver or occurring in dense traffic situations. Here, we describe a hierarchical vision system for the detection and classification of traffic signs on freeways which has been implemented successfully by the mentioned partners. The structure (edges) and color information of an image is used for a detection of traffic sign candidates in the image and their coarse classification. Then regions-of-interest (RoI), each of them containing only one hypothetical traffic sign, are investigated using pixel classification methods. We implemented the entire system on a set of transputers (T800); it has been tested with more than 300 traffic scenes.In this paper, the use of structure information is emphasized and some aspects of its integration into a vision system are demonstrated.them concentrated on road boundary detection as a subgoal of autonomous driving [Franke 1991a, Graefe 1991. Mainly, roads like motorways were analysed which possess clearly visible lane markings . Nevertheless, systems are already implemented and successfully tested for driving not only on highways but also on secondary roads in a robust and reliable manner as shown in [Franke 1991b, Graefe 19911. Up to now, only a few systems are designed for traffic sign detection, one of them is described in [Seitz 19911. Usually, the lane markings are used for a segmentation of the image in order to minimize the analyzing costs.Even less work is done in the field of vehicle rear's identification in stop-and-go situations [Kuhnle 19911.In our group, we have been concentrating on traffic sign detection in scenes obtained from German freeways. Basing on the Hierarchical Structure Code, we adapted already developed tools for its exploitation to the traffic domain and integrated special tools to build a complete vision system for this task. Hierarchical Structure Code IntroductionIn our group, a domain independent image segmentation, the Hierarchical Structure Coding HSC, was developed. Among other applications like object recognition of industrial workpieces [Mertsching 19921, we have been using the HSC for the analysis of traffic scenes as part of the PROMETHEUS project. Within the project, two major goals were pursued: in cooperation with the Institute for Microelectronics in Stuttgart, we built a VLSI-chip for a part of the HSC generation in real-time (25 images/s). Details can be found in [Bilau 19931. Furthermore, we developed fast and domain independent methods for the detection and classification of road boundaries and traffic signs which are presented in this article.For automatic copilot systems, the ability of road tracking, obstacle avoidance and traffic sign recognition is very important; several efforts by a lot of research groups have been made in this field during the last few years. Most of In this chapter, a short survey of the generation ...

Parallel Evaluation of Hierarchical Image Databases

“…All edge structures on resolution level k=0, linking levels n=4..8 are analyzed. The results of our segmentation and classification process have been used as input for pixel classificators to investigate the traffic sign's pictograph[7,20].…”

Road sign interpretation using matching pursuit method

“…Several different techniques have been proposed road sign detection from a sequence of monochromatic [2] or color images [3][4][5][6]. In [2] triangular, octagonal and circular contours, which are likely to represent boundaries of road signs, are selected only among closed edge-contours.…”