In this paper, we describe an approach for human hand motion detection and its reconstruction using an articulated model with hand kinematics constraints. Each finger can be modeled as planar robot arm with 3 joints and 3 links. We assume that first joints connecting palm and each finger configure a rigid form. By this simplification, we have several hand constraints, which reduce the complexity of the estimation process and allow to infer the 3D hand motion and pose in ambiguous situations. The main advantage of the proposed approach is its ability to capture general articulated hand motion with self-occlusion and rotation of the palm. The proposed method is illustrated on a set of examples of a hand motion captured from a monocular image sequence.