An in-vehicle vision system for dangerous situation detection Chiara Corridori, Dimitri Giordani, Paolo Lombardi, Stefano Messelodi, Carla Maria Modena, Michele Zanin {corridori, dimi, lombardi, messelod, modena, mizanin}@itc.it Istituto Trentino di Cultura - Centro per la Ricerca Scientifica e Tecnologica (ITC-irst) Via Sommarive 18, 38050 Povo, Trento, Italy Abstract. This paper presents the on-board computer vision sys- tem under development at ITC-irst inside the automotive project DIPLODOC. The main task of the system is to detect hazard situa- tions in order to alert the driver. The paper describes the status of the obstacle detection and the road recognition modules. 1 Introduction DIPLODOC (DIstributed Processing of LOcal Data for On-line Car services) [1] is a three years project partially funded by the Provincia Autonoma di Trento that started on April, 2002. Three research partners are involved: ITC-irst, CRF (Centro Ricerche FIAT ), and the University of Trento. The goal of the project is to design and develop a system based on a dis- tributed architecture where intelligent vehicles communicate with a remote traf- fic control center. Each vehicle integrates different technologies to provide more comfort and driver safety. Speech recognition and synthesis techniques are used to interact with the user. Computer vision and image understanding are applied to the extraction of traffic parameters and to accident avoidance by detection and recognition of obstacles on the road. Wireless telecommunication is used to send and receive traffic data and route planning information to/from the control center. The practical result of the project will be a simulation of the traffic con- trol center and a demonstrative vehicle (Figure 1) equipped with vision, speech and telecommunication devices. The DIPLODOC architecture has been designed in order to satisfy four ser- vices, defined as follows: – xFCD (eXtended Floating Car Data): the vehicles act as mobile probes for the collection of remote data. Each vehicle transmits its position provided by the satellite positioning system and data coming from the engine control unit and from the on-board vision system. Moreover, still images are sent to a specialized vision module hosted in the center for a deeper analysis. The system at the traffic control center exploits and integrates the information coming from the vehicle fleet in order to deduce the local traffic conditions.