Loop Closure Detection for Visual SLAM Based on Sparse Representation

Simultaneous Localization and Mapping (SLAM) is mainly used to solve the problems of positioning and map construction in the unknown environment of mobile robots. It is the basis for solving the autonomous movement of mobile robots, and loop closure detection can reduce the cumulative error generated by the robot during the movement. When encountering large-scale scenes, the current traditional feature extraction method will have problems: the construction of the dictionary which represents the features will be more complicated, and the calculation of similarity will produce large calculation pressure. For this question, this paper uses the method of sparse representation and principal component analysis to compare the similarity of the sparse feature vectors for loop closure detection. Using Oxford University’s public data set for verification, experiments show that compared with the traditional Bag-of-words model, the proposed algorithm effectively reduces the dimension of image features and retain the image information, and can obtain better accuracy in the loop closure detection.