Determining the Trajectory of Unmanned Aerial Vehicles by a Novel Approach for the Particle Filter

Abstract

There are several applications of Unmanned Aerial Vehicles (UAV), such as environmental monitoring, surveillance, support to the engineering projects, entertainment, among others. The UAV autonomous navigation is one relevant research topic, and a key issue is to estimate the drone position. Here, the signal from a Global Navigation Satellite System (GNSS) is not used to estimate the UAV position. Two approaches are combined to determine the UAV positioning: a computer vision system (edge extraction in images by self-configurating supervised neural networks and correlation with georeferenced images), and visual odometry. The two techniques are employed by a data fusion process by using a new formulation for the particle filter. The particle filter is a Bayesian method, where we are applying the Tsallis’ distribution as the likelihood operator. The Tsallis’ distribution has been employed to derive a generalized statistical mechanics. The filtering process is able to address the fusion data procedure, and to address the uncertanty quatification associated to the UAV trajectory estimation. The methodology was successful in performing the data fusion, generating good results and allowing to compute a confidence interval.