Real-time stereo vision obstacle detection for automotive safety application

Road safety is an important problem that still afflicts our city roads, even though a lot of work has been done into the direction for reducing the amount of accidents. Because of the majority of accidents are due to wrong human behaviour, the safety solutions aimed to improve the intrinsic security of the driving experience, and consequently of the pedestrians, are not enough. Perception systems, that help the driver or even that give the possibility of autonomous action performed by the vehicle, are the next generation automotive safety systems, that potentially could prevent accidents strongly dependent on human errors. The most common and useful perception task for the automotive application is the detection of obstacles. Such task can be achieved, with different characteristics and results, using different kinds of sensors. Our work is focused on the use of the stereo vision in order to realise the detection of obstacles in urban scenario. The use of stereo vision allows to have a cheap solution, that could be potentially mounted on commercial vehicles. Giving particular attention to the performance issue, we obtained a system that well suite urban scenario, with a high operational frame rate on a low cost hardware, that allows to add more high-level modules without the need of more expensive hardware.