The invention relates to a real-time athletic estimating method based on a multiple dimensioned unchanged characteristic, which comprises the steps: (1) a gauss scale space is constructed and a local characteristic point is extracted; (2) a characteristic descriptor of the polar distribution of a rectangular window is constructed; (3) the characteristic point is used for matching and establishing an interframe motion model; and (4) the offset of a current frame output position which corresponds to a window center is calculated. The athletic estimating method has a size, visual angle and rotation adaptive characteristic, can accurately match images with complicated athletic relation, such as translation, rotation, dimension, a certain visual angle change, and the like and has higher real-time performance. The estimating method has better robustness for common phenomena, such as mistiness, noise, and the like in a video, has higher estimated accuracy for arbitrary ruleless complicated athletic parameters and is combined with a motion compensating method based on motion state identification, thus, the image stabilizing requirement of a video image sequence which can be arbitrarily and randomly shot under complex environment can be realized, and the purposes of real-time output and video stabilization can be achieved.