Loop closure detection is very important in visual simultaneous and localization and mapping system (SLAM), which can reduce the accumulated drift. The most successful methods are based on Bag-of-Words model and need a large visual word dictionary. Recently, the VLAD, the state-of-the art compact descriptor, has achieved a great success in large scale image retrieval. In this paper, a novel vlad-based loop closure detection approach is proposed. We compare our method with the state-of-the-art DBoW2 in some open datasets. Our method performs better in accuracy and memory cost.