Improved Reference Key Frame Algorithm

Abstract

<p><strong>Abstract.</strong> The autonomous vehicles, such as wheeled robots and drones, efficiently contribute in the search and rescue operations. Specially for indoor environments, these autonomous vehicles rely on simultaneous localization and mapping approach (SLAM) to construct a map for the unknown environment and simultaneously to estimate the vehicle’s position inside this map. The result of the scan matching process, which is a key step in many of SLAM approaches, has a fundamental role of the accuracy of the map construction. Typically, local and global scan matching approaches, that utilize laser scan rangefinder, suffer from accumulated errors as both approaches are sensitive to previous history. The reference key frame (RKF) algorithm reduces errors accumulation as it decreases the dependency on the accuracy of the previous history. However, the RKF algorithm still suffers; as most of the SLAM approaches, from scale shrinking problem during scanning corridors that exceed the maximum detection range of the laser scan rangefinder. The shrinking in long corridors comes from the unsuccessful estimation of the longitudinal movement from the implemented RKF algorithm and the unavailability of this information from external source as well. This paper proposes an improvement for the RKF algorithm. This is achieved by integrating the outcomes of the optical flow with the RKF algorithm using extended Kalman filter (EKF) to overcome the shrinking problem. The performance of the proposed algorithm is compared with the RKF, iterative closest point (ICP), and Hector SLAM in corridors that exceed the maximum detection range of the laser scan rangefinder.</p>

Multi‐slices navigation approach for unknown 3D environments using micro aerial vehicles

“…Thus, the compact representation is much more proper for search and rescue operations because the utilized MAVs are susceptible to loss during harsh situations. For efficient 3D mapping, many probabilistic approaches such as the voxel occupancy grids (Duffy et al, 1989; Plaza‐Leiva et al., 2015), elevation map (Hadsell et al, 2009; Herbert et al, 1989), multi‐level surface maps (MLS) (Rivadeneyra et al, 2009; Triebel et al., 2006), Octomap (Hornung et al, 2013; Wurm et al, 2010), multi‐volume occupancy grid (MVOG) (Dryanovski et al, 2010), multi‐level occupancy grid (MLOG) (Tian et al., 2016), and Occupancy Elevation Grid (OEG) (Souza & Gonçalves, 2016) are used.…”