Abstract: To address the high-frequency unmanned inspection demands in confined spaces such as substations and cable tunnels, this paper proposes a set of key autonomous drone flight technologies of high safety, low computational power and strong real-time performance. First, a statistical directional prior is constructed, and a constrained enhanced A* algorithm is designed to implement directional pruning in three dimensions. This approach theoretically maintains optimality while reducing the number of search nodes by over 30%. Second, a spherical convex hull safety corridor framework is introduced, characterizing free space via four-dimensional vectors(center+radius). Combined with K-D Tree obstacle queries and adaptive oversampling, this enhances navigation map representation efficiency for unmanned aeriol vehicle(UAV) in confined spaces. Deployed in simulations and real cable tunnel scenarios, the research delivers a safe, efficient technical pathway for resource-constrained UAV platforms to achieve millisecond-level obstacle avoidance in GPS-denied environments.

Key words: unmanned aerial vehicle(UAV), indoor inspection, spherical convex hull, safe flight corridor, GPS-denied, real-time replanning