Navigation with visual measurements in the form of Gaussian Mixture Models

Abstract

Vision-based navigation has emerged as a crucial research area due to the vulnerabilities of the global navigation satellite systems(GNSS). In navigation, dealing with the effect of the variations in lighting, weather, and moving object is important issue. While deep learning techniques have enabled the use of high-level features for navigation, errors such as false positives and false negatives in the visual detector pose significant challenges. This paper presents a novel method for matching images captured by aerial vehicles with images in a database using Gaussian mixture models(GMMs). The features are approximated with GMMs, which can capture the uncertainties and multi-modal nature of the feature distributions. Additionally, we propose a similarity measure based on a data association method to mitigate errors caused by the visual detector. Particle filtering is used to guarantee the accuracy and robustness of vision-based navigation due to severe non-linearity of the measurement model. Numerical simulations show that our proposed method significantly improves the accuracy and robustness of vision-based navigation in challenging environments.