A talk by Gideon Stein about using stereo vision for detecting vehicles and pedestrians. It's an expert system, so this only applies in the narrow context of driving, but still could be quite useful. A combination of stereo correspondence and monocular object detection is used to improve the ranging performance. The fact that they aren't using longer range correspondences with small disparities suggests that they're not using sensor models or doing structure from motion. Possibly I could add this sort of functionality to v4l2stereo if I can get some test images.
Subscribe to:
Post Comments (Atom)
0 comments:
Post a Comment