Object Localization and Ranging Using Stereo Vision for Use on Autonomous Ground Vehicles

摘要

This thesis integrates stereo-vision into existing NPS robot architecture. It demonstrates that image cross correlation can be used to measure ranges as theory predicts. It also demonstrates that objects can be ranged and stored into a database map for later use as common reference points in position determination. Small Unmanned Ground Vehicles (UGV), developed using commercial-off-the-shelf (COTS) technologies are of particular interest for this robotic vision application. To perform their designated missions, these devices require accurate position information. Most devices will determine that position using a Global Positioning System (GPS) receiver; however, the signal is vulnerable to jamming and becomes degraded when not provided a clear view of the sky. Similarly, the error in dead reckoning (DR) systems increases with time if not reset using a known reference. The fusion of stereo vision technology with GPS and DR systems is ideal for use in the design of a command and control module of an unmanned vehicle that is capable of operating autonomously in an environment where traditional position determination loses satellite signals or requires a known reference point to reset uncertainty in position.