Zhiwei Liang and Yanyan Chen
Monocular SLAM, closed-loop detection, visual words, fuzzy C-means, Gaussian mixture models
The closed-loop detection in monocular SLAM for mobile robots is an important issue. To solve this problem efficiently, this paper employed a novel detection algorithm based on a visual dictionary. First, features in each image are described by the SURF method. Then, these visual features were classed into visual words based on a fuzzy C-means algorithm. To consider the statistical distribution (nonlinear and modal characteristics) of the local visual feature vector, we built the probabilistic model of every visual word based on Gaussian mixture models to describe the precise similarities between visual words and corresponding local visual features. As a result, every image can be denoted as a probabilistic vector of visual words and the similarities of any two images can be computed based on a vector inner product. To using the other recent matches, a Bayesian filter method is applied to fuse historical closed-loop detecting information and the current similarities to calculate the posteriori probability distribution of closed-loop hypothesis. In addition, we presented two storage management mechanisms (shallow storage and depth storage) to ensure the real-time performance of our algorithm. The experimental results demonstrate the validity of our approach.
Important Links:
Go Back
