Passive indoor image-aided inertial attitude estimation using a predictive hough transformation

Autonomous navigation of vehicles in non-cooperative environments continues to be a significant challenge for integrated navigation systems due to the unavailability of Global Positioning System (GPS) signals. As a result, passive indoor navigation systems routinely exhibit unbounded growth in navigation errors. In this paper, we seek to stabilize the attitude errors by exploiting commonly-occurring structures within the environment using rigorously coupled image and inertial sensors. Previous work in the literature has demonstrated the value of points on the infinite plane in projective space, also known as vanishing points. These points are invariant to position (translation) changes and are a strong indicator of attitude. If some prior knowledge regarding the location of the vanishing points is available, researchers have shown that these can be used for drift-free attitude estimation. Unfortunately, to our knowledge, research has been limited to sub-optimal implementations based on non-Bayesian techniques and various ad-hoc approaches. We seek to improve this limitation by deriving a statistically-rigorous predictive Hough transformation (PHT), based on a priori attitude information provided by an inertial sensor. The PHT update is then used to correct the inertial sensor using an extended Kalman filter algorithm. The algorithm is tested using a combination of simulation and experimental data. The PHT is shown to improve the robustness of the image-aided navigation algorithm and reduce the effects of outliers on the solution compared to naive approaches. In addition, the navigation solution is show to exhibit drift-free attitude performance using experimental data collected in a typical office environment.