Plenary talk IV: Robust 3-D vision techniques: Algorithms and applications

Summary form only given: For the 3D reconstruction of static and dynamic scenes, we have developed several robust computer vision methods ranging from the image enhancement to the design of novel camera systems. In this talk, we first present a sparse-model based approach to recover the camera response function, which is crucial for many color image processing problems such as high dynamic range imaging and photo-consistent mosaicking. We also present a sensor-fusion based approach to obtain super-resolution (SR) depth images from low-resolution depth images and high-resolution color images. A learning-based optimization framework successfully recovers the SR depth images. The second part of the talk introduces new camera systems to capture the 3D information accurately and efficiently. The camera systems include (i) “camera + depth” fusion camera systems, (ii) a fast bundle adjustment based approach for large-scale datasets, (iii) a novel coded-light photometric stereo for modeling 3-D dynamic scenes. Specifically, we present a hand-held fusion sensor system, consisting of four cameras and two 2D laser scanners, to capture 3D information of large-scale scenes. This new approach allows boosting the advantages of two sensor systems and complements the weakness of the two. As an important application of 3D vision techniques, we demonstrate the robustness of the methods by automatically reconstructing a large-scale environment, such as KAIST campus.