A Preliminary Investigation into an Intelligent Car Headlight Dipping System

Current vision systems for driver assistance are now a huge focus within car manufacturing compaines. Such systems use sensor techniques to detect moving objects, such as pedestrians, or cameras where 360 degree vision is possible using front and rear cameras. The main advantage of using cameras is that visual data is crucial to the detection of moving objects within lanes, the recognition of traffic signs, pedestrians and of particular interest here, car headlights. However, a reliable vision based driver assistance system using even the most sophisticated vision system is extremely difficult as vehicles and objects vary in shape, size and colour and outdoor environments can be very complex. An essential feature of car vision systems is real time recognition of objects and environments. This paper presents a preliminary study into the development of a real-time automatic car headlight dipping using an appropriate combination of image analysis, neural networks topologies and training paradigms.