Autonomous path planning for unmanned aerial vehicles (UAVs) in complex three-dimensional environments requires a balance among search efficiency, obstacle avoidance safety, and trajectory smoothness. However, conventional A* algorithms often suffer from redundant node expansion, insufficient safety awareness, and poor turning performance. To overcome these limitations, this study proposes a hierarchical hybrid planning framework that integrates an improved A* algorithm, particle swarm optimization (PSO), and B-spline trajectory generation. In the global planning stage, a composite cost function is designed by considering path length, safety margin, and turning penalty. Meanwhile, a directional dynamic window and Top-$K$ candidate selection strategy are introduced to reduce invalid expansions and improve search efficiency. In the local refinement stage, key turning regions along the coarse path are identified and optimized using an improved PSO method with adaptive inertia attenuation, reflective boundary handling, and stagnation-triggered reseeding. Finally, B-spline fitting is applied to generate a continuous and executable UAV trajectory. Simulation results show that all compared methods achieved a 100% success rate in the randomized environments. The proposed framework achieved a mean runtime of 20.664 s, compared with 47.108 s for standard A* and 134.666 s for composite-cost A*. Meanwhile, it maintained a comparable path length, indicating robust feasible-path generation, preserved path quality, and acceptable computational feasibility under the tested randomized environments.