Vision-based driver assistant systems are very promising in Intelligent Transportation System (ITS); however, algorithms capable of describing traffic scene images are still very difficult to date. This paper proposes a system which can segment forward-looking road scene image into natural elements and detect front vehicles. First, the scene analysis system deals with scene segmentation and natural object labeling of forward-looking images. By the use of fuzzy Adaptive Resonance Theory (ART) and fuzzy inference techniques, the scene analysis task is accomplished with tolerance to uncertainty, ambiguity, irregularity, and noise existing in the traffic scene images. Secondly, the proposed system can detect the front vehicles and utilize a bounding box shape to further refine the segmentation result. Compared with conventional approaches, the proposed scheme can analyze forward-looking traffic scenes and yield reliable and efficient segmentation results. The validity of the proposed scheme in car detection was verified by field-test experiments. The traffic scene segmentation and front vehicle detection are successful.
Vision-Based Forward-Looking Traffic Scene Analysis Scheme
2007-06-01
532599 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
Vision-Based Forward-Looking Traffic Scene Analysis Scheme
British Library Conference Proceedings | 2007
|Deep learning serves traffic safety analysis: A forward‐looking review
Wiley | 2023
|Looking Backward, Looking Forward
British Library Conference Proceedings | 2003
|