Addressing the path planning challenge for Underwater Tracked Vehicles (UTVs) in complex marine environments, this study proposes a multi-constraint optimization method integrating seabed terrain mechanics and dynamic current fields. A physically realistic three-dimensional marine environment is established by constructing a variable subsidence coefficient seabed soil model based on normalized terrain data and a Lamb-Oseen vortex field model. The core algorithm employs an improved adaptive Particle Swarm Optimization (PSO) method, innovatively introducing a terrain-fitting constraint mechanism and velocity mirroring boundary processing technique. Learning parameters are dynamically adjusted based on particle aggregation factors to balance global exploration and local exploitation. The path evaluation function couples dual metrics of energy consumption and safety: the energy model integrates seabed soil bulldozing resistance and ocean current viscous resistance, while the safety module achieves static obstacle avoidance. Experimental results demonstrate that the generated paths strictly adhere to terrain constraints, effectively utilize favorable currents to reduce energy consumption, providing a safe and energy-efficient path planning solution for autonomous underwater tracked vehicles.