An important aspect of occupant protection systems like air bags and pretensioners is the time needed for their activation. Current systems are based on electronic crash sensors which respond to sudden decelerations resulting from an impact. Those systems need time to react. The loss of time results from the fact that the progress of deceleration has to be evaluated to come to the final decision to trigger the on board safety systems. Furthermore, they do not give precise information about the type of the collision like pedestrian crash, car crash and so on. Especially accidents with pedestrians are actually in the focus of the legislators all around the world. On board pedestrian protection systems are still object of research, but nevertheless they will come and they will need electronic object sensing. This paper presents a computer vision based approach to overcome these limitations. Pre-crash detection and the recognition of possible collision candidates is done by means of an on board camera system. The image is scanned for hypotheses of different types of road users by a hyper permutation network, a new kind of neural network architecture which was developed by the DaimlerChrysler research. This scan is performed in real time. After a fast clustering procedure the algorithm provides a list of boxed object candidates which are further processed by a standard classifier stage to get the type of the object. Current object types are 'truck', 'car', 'human' and 'garbage'. The detection and recognition system is designed in a flexible way. If a second camera is available in the vehicle, the estimation of the closing velocity and the object tracking is done by using both cameras as a stereo camera system. On the other hand, if there is a suitable microwave sensor, the output of this device can be used to get the above information and only one camera is needed for the classification task.
Detection and classification of potential collision candidates in traffic scenes
Detektion und Klassifizierung potentieller Verkehrunfallsituationen
11 Seiten, 7 Bilder, 4 Tabellen
Conference paper
English
Detection and classification of potential collision candidates in traffic scenes
Automotive engineering | 2000
|Detection and Classification of Potential Collision Candidates in Traffic Scenes
British Library Conference Proceedings | 2000
|Efficient pedestrian detection in urban traffic scenes
TIBKAT | 2015
|Monitoring crowded traffic scenes
IEEE | 2002
|