Rapid 3D visualization of indoor scenes using 3D occupancy grid isosurfaces

In many mobile robotics applications involving exploration of unknown environments, it would be extremely useful to provide human operators with a real-time 3D visualization of the environment the robot is exploring. Although a great deal of progress has been made in the separate fields of photorealistic structure from motion and realtime vision-based robot localization and mapping (SLAM), the ultimate goal of real-time 3D visualization of the environment a robot is exploring has yet to be realized. In this paper, we present a simple and efficient incremental algorithm for 3D modeling amenable to realtime implementation. The algorithm creates a texture-mapped polygonal mesh model of the environment from a monocular video feed or sequence of images. The key to the algorithm's simplicity and efficiency is the use of the isosurface of a coarse 3D occupancy grid that is incrementally updated as new images arrive. The isosurface-based reconstruction provides low metric accuracy but helps to filter measurement noise and allows rapid construction of a 3D visualization. We demonstrate the practicality and effectiveness of the algorithm by using it to generate an OpenGL model of a real indoor environment.