[Purposes] In response to unmanned aerial vehicle (UAV) localization failure in satellite-denied environments, threats from dynamic obstacles, and insufficient path planning safety and real-time performance caused by complex terrains, a UAV path planning method based on hierarchical risk perception was proposed. [Methods] Based on hardware configuration, a ''global guidance-local optimization'' hierarchical architecture was established. At the global layer, an improved D*-Lite algorithm was employed. The obstacle distance risk kernel function and UAV kinematic constraints were integrated to generate a coarse-grained risk perception map and output a safe corridor. At the local layer, a dynamic-biased Informed-RRT* algorithm was utilized to adaptively adjust the sampling probability distribution within the safe corridor, avoid high-risk grids, and achieve real-time dynamic obstacle avoidance as well as path smoothing optimization. [Findings] According to the simulation experiments, in complex scenarios with an obstacle density of 25%, the proposed algorithm achieves a planning success rate of 96.8%, 22.4% and 8.3% higher than the traditional D* algorithm and the Informed-RRT* algorithm, respectively. The planning path length ratio is reduced to 1.14, the average path risk value decreases to 0.21, and the average planning time per instance is only 87 ms. Real-world flight test further confirms that the system can effectively avoid dynamic obstacles in real time within a distance of 0.5 m, meeting requirements of centimeter-level positioning accuracy and 22-min endurance time. [Conclusions] The path planning method based on hierarchical risk perception significantly enhances UAV navigation safety, path quality, and real-time performance in satellite-denied environments, providing a safe and efficient autonomous flight solution for complex scenarios such as post-disaster rescue.