The aim of this paper is to investigate the 3D (three dimensional) environment reconstruction for mobile robot without using GPS (global positioning system). Based on the real-time pose estimation by using the stereo, we used an inlier selection approach developed from the max-clique to improve the performance of RANSAC, then computed the 3D pose with a robust pose estimation algorithm. A 3D laser system to achieve an indoor 3D point cloud data collection was designed and developed. The multi-view point cloud was further merged based on the reliable pose information obtained from the visual odometer, which is able to provide an effective large scale 3D environment reconstruction map. The proposed approach was simulated and tested on the SmartROB platform in indoor rooms, corridors and halls. The results has represented and verified the effectiveness and practicability of the approach.