Visual Exploration with UAVs: Solving the Next-Best-View Problem with Limited A Priori Information

Little research has investigated autonomous methods for exploration and mapping by UAVs that operate without active depth sensors or prior directions. This work presents a novel approach to the Next-Best-View (NBV) problem for UAVs exploring unknown objects solely equipped with a camera and operating within a constrained flight time. The vehicle first performs a predefined search routine to locate and obtain several initial views of the target. Subsequent view selection is informed by sparse point clouds generated in real time and leverages the You Only Look Once (YOLO) framework to dynamically define a 3D region of interest (ROI) based on AI inference. The search algorithm identifies gaps within the ROI, allowing the UAV to autonomously prioritize areas requiring further inspection. This method provides an efficient exploration strategy that is more flexible than systems with predefined search trajectories. This paper presents analysis and simulation of the proposed method, supported by initial results.