Hybrid Depth Imaging

We propose to create a three camera system consisting of two conventional cameras and one Canesta 3D imaging sensor. The two conventional cameras will operate as a stereo pair, with the Canesta sensor providing an estimate of the disparity, limiting the computation required to find a strong match, resolving ambiguous matches, and providing information about texture-less image regions. This creates a depth sensor with greater resolution than the Canesta sensor and improved robustness over classical stereo. We intend to experiment with the hybrid system in robot navigation, head-pose estimation and object recognition.