Abstract: A underground image tracking algorithm is developed by way of a fusion done between a commonly-used mean square image correlator and Kalman filter, based on Bayes rule. With the fusion of both the information from the MSD correlator and Kalman filter, the improved algorithm can enhance the information feedback between them and its tracking performance and robustness so as to minimize the out-of-control possibility. Furthermore, the improved algorithm also incorporate the statistical characteristics of noise to improve noise suppression ability. The theoretical and practical results show that the images acquired from the algorithm are much more real and accurate than the correlative algorithms.