In this paper, two methods for hand tracking and and online hand guesture identification is proposed by using the combination of Leap Motion and MYO armband. With the proposed methods, We have improved the measurement accuracy of the palm direction and solved the problem of insufficient accuracy when the palm are at the limit of the measurment range. We use the Kalman filter algorithm and the neural network classification method to process and analyze the data measured by Leap Motion and MYO, so that the tracking of the operator's hand gesture is more accurate and robust even when the hand is at positions close to the measurement limit of one single sensor. The improved hand tracking method can be used for robotic control, teaching by demonstration or teleoperation. The effectiveness of the proposed methods have been demonstrated through comparative experiments.