This paper proposes HTCplanner, a real-time hierarchical path planning framework for bipedal robots, designed to improve adaptability in complex terrains through integrated traversability and crossability analysis. The method combines PRM sampling with graph search to evaluate edge feasibility based on bipedal kinematic constraints, enabling dynamic optimization of global centroid paths and local footstep sequences. In the global planning stage, a 2.5D elevation map and traversability-aware cost function generate collision-free paths, while local planning employs coupled foot state sampling guided by the global trajectory, reducing computational complexity and improving coordination between bipedal steps. Key contributions include: (i) a unified hierarchical architecture that seamlessly bridges global and local planning, (ii) an optimization strategy for bidirectional footstep sequences through state coupling, and (iii) comprehensive validation in both simulated and real-world scenarios, demonstrating significant improvements in planning efficiency and obstacle negotiation compared to conventional approaches. The framework addresses critical challenges in real-time navigation for bipedal robots, particularly in unstructured environments, with potential applications in hazardous terrain exploration. The implementation is open-source at: https://github.com/chenfu-user/HTCplanner .