Enhancing UAV Search under Occlusion using Next Best View Planning

TL;DR

This work tackles UAV-assisted search in occluded environments by formulating NBV planning as a nonconvex optimization problem evaluated with two heuristics: Visibility Fitness $J_v$ and Geometry Fitness $J_d$. The authors implement an Evolutionary Algorithm (via the DEAP library) to optimize candidate camera poses, with $J_v$ leveraging ray-traced visibility of mesh vertices and a tunable weighting scheme, and $J_d$ using a D-optimality criterion grounded in photogrammetric linearization of the co-linearity equations. Across simulations and real-world experiments, $J_v$ consistently achieves higher hidden-object coverage (≈90% in simulations) and superior canopy-area visibility, while $J_d$ increases per-object pixel intensity but at the cost of broader coverage. The results demonstrate that NBV-guided UAV search can substantially improve detection in occluded forests, guiding future autonomous SAR deployments in dense environments.

Abstract

Search and rescue missions are often critical following sudden natural disasters or in high-risk environmental situations. The most challenging search and rescue missions involve difficult-to-access terrains, such as dense forests with high occlusion. Deploying unmanned aerial vehicles for exploration can significantly enhance search effectiveness, facilitate access to challenging environments, and reduce search time. However, in dense forests, the effectiveness of unmanned aerial vehicles depends on their ability to capture clear views of the ground, necessitating a robust search strategy to optimize camera positioning and perspective. This work presents an optimized planning strategy and an efficient algorithm for the next best view problem in occluded environments. Two novel optimization heuristics, a geometry heuristic, and a visibility heuristic, are proposed to enhance search performance by selecting optimal camera viewpoints. Comparative evaluations in both simulated and real-world settings reveal that the visibility heuristic achieves greater performance, identifying over 90% of hidden objects in simulated forests and offering 10% better detection rates than the geometry heuristic. Additionally, real-world experiments demonstrate that the visibility heuristic provides better coverage under the canopy, highlighting its potential for improving search and rescue missions in occluded environments.

Figures (8)

• Figure 1: The scaling function for the weights $\alpha_i$ in \ref{['equ:scalin_fucntion']} versus vertex visibility.
• Figure 2: Workfow of simulation experiments to evaluate the proposed NBV planning.
• Figure 3: Example of the outcome of a simulation run. The picture shows the randomized forest scene with one hidden manikin marked in red. The 36 blue cameras represent the initial views, while the 20 green cameras represent planned NBV according to Visibility Fitness. On the left, a section of the top-down camera illustrates how the manikin is obstructed by the trees. In contrast, on the right, one of the planned camera views shows a clear visibility of the manikin.
• Figure 4: Convergence of evolutionary algorithm for the $J_v$ fitness plotted over the number of generations. The blue curve represents the maximum achieved fitness in the population for a particular generation, while the red curve shows the average in the population.
• Figure 5: The curves show how many of the 100 hidden manikins are visible with respect to the placed camera views. For the blue curve, the Visibility Fitness was used during the view optimization, and for the red curve the Geometry Fitness, respectively. The curves are averaged over 18 simulation runs with randomized environments. The transparent area marks the minimum and maximum values for the 18 runs.