Path planning in complex environments is fundamental to ensuring safe UAV operations. A global optimization method was proposed that comprehensively considered obstacle avoidance capability, flight efficiency, and stability constraints. The cylindrical envelope method was employed to standardize environmental data and obstacles, and the impact of UAV attitude variations on motion stability and flight efficiency was analyzed. Three optimization objectives—efficiency, obstacle avoidance, and stability—were established, expanding the optimization space from traditional one-dimensional or two-dimensional to three-dimensional. A reference-point-based non-dominated sorting genetic algorithm III (NSGA-III) was introduced to accelerate the search process in high-dimensional space, achieving approximately 49% faster convergence compared to the improved A* algorithm. A potential risk point detection mechanism was designed for obstacle avoidance, and a quintic B-spline was used to perform secondary optimization on the initial optimal path, enhancing the geometric characteristics and smoothness of the UAV trajectory. As a result, flight stability and efficiency were improved by 66.7% and 25%, respectively. The effectiveness of NSGA-III in trajectory planning was validated through simulations and experiments.