Coupling Vanishing Point Tracking with Inertial Navigation to Estimate Attitude in a Structured Environment

This research aims to obtain accurate and stable estimates of a vehicle's attitude by coupling consumer-grade inertial and optical sensors. This goal is pursued by first modeling both inertial and optical sensors and then developing a technique for identifying vanishing points in perspective images of a structured environment. The inertial and optical processes are then coupled to enable each one to aid the other. The vanishing point measurements are combined with the inertial data in an extended Kalman filter to produce overall attitude estimates. This technique is experimentally demonstrated in an indoor corridor setting using a motion profile designed to simulate flight. Through comparison with a tactical-grade inertial sensor, the combined consumer-grade inertial and optical data are shown to produce a stable attitude solution accurate to within 1.5 degrees. A measurement bias is manifested which degrades the accuracy by up to another 2.5 degrees.