With the wide application of simultaneous localization and mapping(SLAM) in various fields in recent years, the algorithms are more and more mature and the system tends to be stable. However, many problems still need to be solved urgently. The accuracy and efficiency of the SLAM system have always been the major problems, focusing on the difficulties of the key frame extraction, hard to determine the loop location in the loop detection process, poor tracking performance and so on. In this paper, we propose a key frame extraction algorithm based on rotation degree and a loop detection algorithm based on the historical differential loop model. The experiments show that the efficiency of loop detection and the success rate of relocalization have been significantly improved. The robustness of the system is stronger and the tracking performance is better by using the improved algorithm.