Inverse kinematics, object localization and manipulation are essential for service robots with manipulators to achieve various human-like tasks. Binocular vision is employed to interact with the environment in which the service robot works. This paper presents an approach for binocular based humanoid manipulation in a service robot system. An inverse kinematic solver is proposed to find all joint angles for a given position of the effectors on the manipulator. The target object is recognized according to segmented colors, and the 3D position computed using the stereo vision system. Having obtained the target position, the manipulator performs a blind grasp. Experimental results show effectiveness of the proposed methods.