Interactive scene modeling from dense color and sparse depth
暂无分享,去创建一个
Scene modeling is a major bottleneck for computer graphics applications. A promising approach is to build the scene model from acquired color and depth data. Systems that build highquality models are slow (e.g. depth from stereo, laser rangefinding), require expensive and bulky equipment (e.g. laser rangefinding), and require manual intervention (e.g. image-based editing, view morphing, tour into the picture). Systems that do not suffer from these disadvantages build simplistic scene models (e.g. color panoramas) that must be viewed from preferred locations. We present a fast, easy to use, and inexpensive modeling system that builds scene models that support realistic interactive rendering from a wide range of viewing locations. Our system acquires video frames augmented with sparse depth samples. The frames are registered and merged into an evolving model at the rate of five frames per second. The model is displayed continually to provide immediate feedback to the operator. We illustrate the system on an office scene; we obtain meshes of 400,000 triangles in 15 minutes.