In order to stabilize the output video of camera system in good time, a real-time electronic image stabilizing algorithm based on feature matching is presented. The improved Kanade-Lucas-Tomasi (KLT) algorithm is applied to feature extracting, and a new feature matching algorithm based on projection mean is put forward. Considering the reliability of stabilizing results, a feature validation test method is also given. Besides, based on the high real-time demand for motion filtering, the recursive Kalman filtering is applied in intentional motion parameter estimation. As experiments show, the algorithm just takes 24.7ms to stabilize a single image on PC, fast enough to satisfy the real-time demand, and it has good stabilizing results.