Tracking-Learning-Detection (TLD) is a kind of long-term visual tracking algorithm which receiveds wide attention in recent years. In order to improve the running speed of this algorithm, a novel algorithm named Accelerated TLD (ATLD) is proposed in this paper. Two aspects of improvements were made in original TLD algorithm. The improvement includes as follows: using a grey prediction model in the detection module for estimating the location of the target and setting a detection area; applying an image indexing method based on normalized cross correlation (NCC) distance to manage the positive and negative sample set. And on this basis, the multiple targets tracking algorithm is realized. Through experiments, the ATLD algorithm, the original TLD algorithm and other two recent improved TLD algorithm are compared. The experimental results show that the ATLD algorithm runs faster on the premise of ensuring the accuracy.