In this paper, we propose a real-time measuring method of a moving three dimensional object. We use a single camera image and an extended Kalman filter to track the object. Assuming that the velocity of the object is constant for a sufficiently small time interval, we construct a descrete-time extended Kalman filter on a simple motion model. The Kalman filter predicts the object position at the next sampling step from the projected position and area of the object in the current sampling image. Although we use a simple model, we can increase the estimation accuracy with an appropriate choice of the covariance matrix reflecting the effect of the system noises. We derive the discrete-time covariance matrix directly from the continuous-time model. It is pointed out that the covariance matrix usually contains additional correlation terms which cannot be ignored for a large sampling period. By simulation and experiment, we show the effectiveness of the proposed method for pendulum motion estimation.