Fast 3D reconstruction for small autonomous robots

The reconstruction of the 3D structure of a static environment is a powerful capability applicable to robot navigation. We present a novel approach to determine the 3D positions of a sparse set of points tracked by a small autonomous robot in a static environment. Using noisy feature tracking and a rough estimate of the essential matrix, our method has a relatively low computational cost and runs at frame rate (15 fps) on a small FPGA-based robot platform. The working implementation demonstrates the power of FPGA hardware/software co-design in small robotics applications. In addition, we describe our implementation of a real-time image derivative computation system for use by our vision algorithm. We present results of using our method to navigate a test environment.