Abstract
Small-Unmanned Aerial Vehicles (S-UAVs) provide significant advantages in several civilian applications such as search-and-rescue operations in disaster regions, anti-poaching operations, and policing for civil security. In essence, they provide a low-cost alternative for obtaining a better situational awareness through a bird’s-eye view of unreachable locations. However, their effective coverage area is starkly reduced by the presence of occlusions in the field-of-view(FOV) of the sensor. In this work, to obtain a quick and effective aerial surveillance, three path planning strategies are developed that enhancesground/target visibility.
The first path planning approach improves the ground visibility over large cluttered environments with vegetation. Visibility occlusions are considered in two forms: partial occlusions due to vegetation and complete occlusions due to buildings and terrains. The gradual reduction in visibility along the line-of-sight(LOS) due to foliage in a forested region is modeled probabilistically using the crown cover density of that region. To obtain near-uniform visibility of the ground points, the waypoints (also the imaging points) are set as the generator points of a Centroidal Voronoi Tessellation(CVT). The CVT is computed with a combination of the forest crown cover density and the topographical profile as the density distribution function. The Dubin’s flight path through these waypoints is solved by an improved spiral-alternating algorithm. Visibility with the proposed method is computed for: (1) a synthetically generated forest terrain, and (2) actual satellite tree cover data & digital elevation model. It is then compared with the visibility from regular grid based observation points.
The second path planning approach optimizes both the duration of a target visibility and path length in an urban environment. For an urban region modelled in the form of 3D occupancy grids, determination of from-point visibility is computationally very expensive. In this work, we utilize a parallel Fast-marching method(FMM) to compute the 3D visibility for real-time applications. This work also attempts to standardize the selection of urban models for simulation studies by different researchers. A Visibility-Based Fast-Marching field is constructed to function as the cost field for the path planning. A 2-step finite horizon local path planner is also proposed that incorporates the kinematic constraints of the UAV and collision avoidance. This algorithm is compared with a simple shortest path planner.
Another path planning strategy is proposed for searching and tracking a moving target in urban environments. An algorithm to determine the camera footprint for a given view direction from the S-UAV is developed. The target detection is achieved through color segmentation and the target is localized in the world coordinate system with photogrammetry. It is assumed that the target’s position is not known to the S-UAV and it estimates the target location using particle filters. We also developed the k-medoids clustering method for selecting the best goal points for path planning and best view directions for the two-axis gimballed camera.
Highly realistic simulations are performed using the Robot Operating System(ROS) and the Gazebo platform. The results show the effectiveness of the proposed path planning methods in enhancing the visibility of targets in cluttered environments (buildings and trees).
Concepts Used: