Micro Aerial Vehicles (MAV) Assured Navigation in Search and Rescue Missions Robust Localization, Mapping and Detection

Abstract

This Master's Thesis describes the developments on robust localization, mapping and detection algorithms for Micro Aerial Vehicles (MAVs). The localization method proposes a seamless indoor-outdoor multi-sensor architecture. This algorithm is capable of using all or a subset of its sensor inputs to determine a platform's position, velocity and attitude (PVA). It relies on the Inertial Measurement Unit as the core sensor and monitors the status and observability of the secondary sensors to select the most optimum estimator strategy for each situation. Furthermore, it ensures a smooth transition between filters structures. This document also describes the integration mechanism for a set of common sensors such as GNSS receivers, laser scanners and stereo and mono cameras. The mapping algorithm provides a fully automated fast aerial mapping pipeline. It speeds up the process by pre-selecting the images using the flight plan and the onboard localization. Furthermore, it relies on Structure from Motion (SfM) techniques to produce an optimized 3D reconstruction of camera locations and sparse scene geometry. These outputs are used to compute the perspective transformations that project the raw images on the ground and produce a geo-referenced map. Finally, these maps are fused with other domains in a collaborative UGV and UAV mapping algorithms. The real-time aerial detection of victims is based on a thermal camera. The algorithm is composed by three steps. Firstly, a normalization of the image is performed to get rid of the background and to extract the regions of interest. Later the victim detection and tracking steps produce the real-time geo-referenced locations of the detections. The thesis also proposes the concept of a MAV Copilot, a payload composed by a set of sensors and algorithm the enhances the capabilities of any commercial MAV. To develop and validate these contributions, a prototype of a search and rescue MAV and the Copilot has been developed. These developments have been validated in three large-scale demonstrations of search and rescue operations in the context of the European project ICARUS: a shipwreck in Lisbon (Portugal), an earthquake in Marche (Belgium), and the Fukushima nuclear disaster in the euRathlon 2015 competition in Piombino (Italy).