Panoramic stitching for driver assistance and applications to motion saliency-based risk analysis

Cameras can provide the capability for surround vehicle analysis and intuitive visualization for driver support in critical situations. As salient objects can be distributed anywhere around the vehicle, a wide view of the surround has many benefits for studying driver behavior in a holistic manner, as well as developing effective driver assistance systems. In this work, we are concerned with the hardware setup and calibration needed to provide such stitched views. In particular, views with large translation differences, many moving objects, and large changes in brightness are handled. We show the qualitative effects of calibration scene and camera orientation on common stitching algorithms. Finally, we analyze the stitched view for salient objects to detect critical events.