6D Vision Goes Fisheye for Intersection Assistance

For automotive applications, 3D perception of the car's surroundings is crucial, both for driver assistance and for safety systems. In addition, a large field of view is required for future applications such as intersection assistance. A popular option to obtain 3D measurements and detect objects is to use two cameras (stereo vision). Applications based on object detection range from  Adaptive Cruise Control, Collision Mitigation Systems to Intersection Assistance Systems. Stereo vision with conventional cameras only delivers a limited field of view. We extend the field of view using fisheye lenses. In this contribution a detailed description of a fisheye stereo system for object detection is given, including the steps calibration, rectification, and stereo computation. Differences to traditional stereo systems are pointed out and the accuracy for an experimental setup is evaluated. In addition, the fisheye stereo approach is combined with optical flow measurements to obtain position and velocity data simultaneously. This so-called 6D Vision approach is extended to stereodata obtained with fisheye lenses. Results for intersection scenes are presented.