Grid Map based Free Space Estimation using Stereo Vision

This contribution proposes a temporally filtered free space estimation method for autonomous driving using dense disparity images from stereo vision. Urban environments feature complex surroundings in which the free space is limited by large and relatively flat obstacles (e.g. cars and curbs). Free space methods relying on single frame measurements suffer from sensor noise and depth artifacts, leading to large deviations from the ground truth free space. We meet this challenge by temporally filtering the occupancy of height and orientation features in a probabilistic occupancy grid. In the following the free space boundaries are estimated from the origin of the sensor in the occupancy grid. In contrast to existing methods our approach allows us to detect static free space boundaries which are not in the current sensor’s field of view. The proposed approach is evaluated on different urban scenes and compared to a state of the art free space method.