Through our investigation of traditional A* algorithms and their variants, we identified a critical safety flaw: conventional path planning tends to excessively follow obstacle contours, significantly increasing collision risks for UAV navigation. To address this issue, we developed a probabilistic collision model incorporating grid-based obstacle mapping and UAV dynamic instability parameters. Building upon this model, we propose an enhanced A* algorithm that integrates both adaptive environmental heuristics and global correction factors. Experimental results demonstrate remarkable safety improvements: an 83.5% increase in average path clearance distance, complete elimination of obstacle-hugging behavior, and substantial mitigation of collision risks from 24.94 percent to 2.89%