Information theory is combined with the Rao-Blackwellised particle filter (RBPF) for mobile robot simultaneous localization and mapping (SLAM). The new version of SLAM is termed active SLAM. This paper addresses the problem of maximizing the accuracy of the building map during active exploration by adaptively selecting control actions that maximize localization accuracy. The map information is maximized by simultaneously maximizing the expected mutual information gain on the 3D occupancy grid map minimizing the uncertainty of the robot pose and map landmarks uncertainty in the SLAM process. Monocular vision mounted on the robot tracks Scale Invariant Feature Transform (SIFT) feature. The matching for multi-dimension SIFT features is implemented with a KD-Tree in the time cost of O(log2N). Experiment results on Pioneer robot in a real indoor environment show the practicality and efficiency of our proposed method.