According to measurement error which is brought by collision and extrusion between the bottles and imaging distortion from the camera during the motion of the bottles, we proposed an adaptive measurement covariance of discrete kalman filter method. By dynamically adjusting measurement covariance, the method improves the tracking precision of the target motion as the basic condition of grab. The solution is verified by the experiment. Under image acquisition rate of 20 frame/s and maximum operating speed more than 0.5 m/s, the algorithm can effectively overcome the influence of measurement noise and data loss situation. Besides, the proposed method also can give an excellent tracking result when the moving radius of the bottle is varying.