A precise perception of the environment of the ego vehicle's vicinity forms the basis for many driver assistant systems. It is furthermore a crucial function for autonomous driving.
Therefore we are currently researching various methods of computing a representation of the environment surrounding the vehicle from stereoscopic image sequences.
Dense depth maps computed from stereo images are transformed into the grid coordinate system representing the vicinity of the ego vehicle. Different layers contained in the grid are updated sequentially to obtain the free and occupied space surrounding the vehicle.
Two dimensional maps suffer from the inability to distinguish between objects of different height. This is mainly due to the projection of the three-dimensional world into a two-dimensional plane. Therefore a three dimensional representation is sought.