The upper limbs of the humanoid robot consist primarily of an arm-hand system for executing tasks and a sensing system for perception. The anthropomorphic grasping of the upper limbs holds significant research and practical implications, which involves precise perception and control of objects in the environment, enabling humanoid robots to execute intricate motions and tasks effectively. In this paper, an upper-limb motion planning approach is presented for realizing human-like dexterous operation. The camera installed on the humanoid robot provides visual perception, capturing depth images and point cloud data of the external environment for the purpose of object detection and recognition. The arm-hand system is the task execution unit of the humanoid robot. Inspired from human movement characteristics, the dynamic movement primitives are employed to generate collision-free continuous trajectories for the manipulators. Moreover, unlike conventional end-effector such as the parallel grippers, the robotic arms are equipped with five-finger dexterous hands at its end to perform humanoid grasping tasks. Stable grasping is facilitated by force-closure gesture generation and position-based impedance control of the dexterous hand. Various experiments are conducted to demonstrate impressive operational flexibility of the humanoid robot.