Cooperation-Aided Accurate UAV-Based LiDAR Mapping: Experimental Assessment

This paper presents results of an experimental activity conducted to prove the feasibility of using cooperation among Unmanned Aerial Vehicles to improve the geolocalization accuracy in Light Detection and Ranging (LiDAR)-based mapping applications. A chief-deputies architecture is considered, where the navigation state of the chief platform (which is embarking the LiDAR payload) is estimated by fusing measurements of its onboard Inertial Measurement Unit (IMU) and Global Navigation Satellite System (GNSS) receiver, with IMU-independent attitude measurements obtained by exploiting cooperation with the deputy platforms. Specifically, the chief attitude is computed by tightly integrating visual- and Carrier-phase Differential GNSS (CDGNSS) measurements of the chief-deputies directions in body and local coordinates, respectively, in the correction step of an extended Kalman filter. An experimental flight test has been conducted in a powerline mapping scenario to assess the performance of this cooperative technique in providing accurately geolocalized data. The experimental setup includes the following items: one customized DJI M300 platform (acting as chief) equipped with a Livox MID-40 LiDAR (mapping payload) and a fisheye camera (cooperative visual sensor); and two customized DJI M100 platforms acting as deputies. The three platforms are also equipped with GNSS receiver with raw data capability to enable CDGNSS processing. Results demonstrate an ability of reconstructing powerlines with a 0.42m Root Mean Square dispersion of the geolocalized points.

I. Introduction

The advances in Unmanned Aerial Vehicles (UAV) technology have greatly changed the way of executing several tasks, reducing the setup and mission time, as well as the cost and the human risk. Mapping is one of the applications which mostly benefits from being carried out by UAVs. Indeed, these platforms allow saving time and achieving accurate results, while also enabling missions in sites hardly reachable with other manners. Different electro-optical sensors can be used as mapping payloads, including Light Detection and Ranging systems (LiDAR) [1] and Cameras [2], [3]. While camera images require triangulation of visual features to infer depth measurements and perform 3D map reconstruction, LiDAR point clouds directly provide 3D information without further processing. Several works in the open literature [4]–[7] have used the latter as a powerful tool to perform map reconstruction due to its associated high density and accuracy in retrieving points information.