To address the challenges of unsmooth paths, low computational efficiency, and local optima in real-time obstacle avoidance path planning for unmanned aerial vehicles (UAVs) operating in complex dynamic environments, this study proposes a integrated optimization approach integrating Bézier curves and dynamic sliding windows. First, differentiable paths are constructed using segmented quadratic Bézier curves, and a weighted multi-objective optimization model is established through gradient descent to simultaneously minimize path length and flight time. Second, a dynamic sliding windows mechanism is designed to decompose the global path into local rolling optimization sequences, enabling real-time obstacle avoidance while maintaining computational efficiency. Additionally, UAV motion constraints are incorporated to ensure the generated paths align with practical flight capabilities.Simulation experiments demonstrate that, in scenarios containing both static and dynamic obstacles, the proposed method outperforms traditional dynamic window approaches in path smoothness, computational efficiency, and real-time responsiveness. The generated paths also approximate the global optimum in length. This study offers a practical solution for real-time path planning in dynamic environments, balancing smoothness, safety, and computational efficiency.