Loop Closure Detection based on Image Covariance Matrix Matching for Visual SLAM

Loop closure detection is an indispensable part of visual simultaneous location and mapping (SLAM). Correct detection of loop closure can help mobile robot to reduce the problem of cumulative pose drift. At present, the main method for detecting visual SLAM loop closure is the bag of words (BoW) model, but it lacks the spatial distribution information of local features of the image, and the scale will become larger and larger with the increase of data, resulting in the slow operation speed. In order to solve these problems, the image histogram and the key region covariance matrix matching method are used to visually detect the loop closure combined with the global and local image features. In this paper, three different place recognition techniques are studied: histogram only, image covariance matrix matching (ICMM) and cluster loop. Experiments on real datasets show that the proposed method of detecting the loop closure is better than the traditional methods in detecting accuracy and recalling rate, which also improves the operation effect of the SLAM algorithm.