In recent years, ground-based autonomous mobile robots have been playing an increasingly important role in public security as well as reconnaissance, patrolling and resupply tasks in militarized scenarios. However, traditional global path planning algorithms often find it difficult to simultaneously balance planning speed and path quality when facing battlefield terrains with dramatic ups and downs, complex obstacles, and limited communication sensing. In this paper, we propose the Connect-RRT* algorithm with slope barriers and local circular domain sampling constraints. Firstly, we organically combine the advantages of classical RRT, RRT* and Connected-RRT algorithms, on which we propose the Connect-RRT* algorithm, which is capable of both fast path finding and path optimization. Then we introduce elevation information and the maximum feasible angle of the robot, as well as local circular domain sampling based on the Connected-RRT* algorithm for applications in militarized 3D unstructured scenes. Finally, 2D and 3D comparison experiments and sensitivity analysis of sampling parameters are conducted with classical RRT, Connected-RRT, and RRT* algorithms in a real complex unstructured environment. The experimental results show that the proposed method significantly outperforms the control algorithms in terms of planning time and path quality, which verifies its practical value in military unstructured scenarios.