Given the characteristics of dense obstacles, complex structures, and dynamic changes in urban environments, it is necessary to evaluate the performance of different perception and recognition technologies and path planning algorithms, and develop a set of optimized design methods for autonomous obstacle avoidance by drones to ensure their safe operation in urban environments. Firstly, the principles, performance, and applicable scenarios of three types of perception and recognition technologies, namely visual sensors, millimeter-wave radar, and lidar, were analyzed and compared. Subsequently, A*, RRT, and the ant colony algorithm were selected as representative path planning methods. An obstacle environment was constructed in MATLAB, and a comprehensive evaluation function was designed to conduct comparative simulations of the three algorithms in a static environment. Furthermore, the RRT algorithm was simulated in a dynamic obstacle scenario. Through static environment simulations, the feasibility of the three algorithms was verified, with the A* algorithm demonstrating the best overall performance and the RRT algorithm having the shortest computation time. Dynamic environment simulations further verify the real-time performance of the RRT algorithm. For urban drone obstacle avoidance, multi-sensor fusion is required to improve the accuracy of environmental perception, and a hierarchical path planning strategy is adopted to balance performance and efficiency.