In this study, we aim to achieve an autonomous locomotion of the mobile robot in the unknown indoor environment. In particular, we propose a novel landmark construction and detection method. The landmark is generated by combined image and shape features. In the landmark detection step, we represent the new matching with confidences defined automatically. The effectiveness of the proposed landmark is verified by experiment. Moreover, we introduce the novel landmark based graph SLAM. In our method, that landmark detection is performed on each node. Then, if the robot find the landmark it found once, local loop closure is generated and optimization is performed. The advantage of this method is that we can perform graph optimization before find loop-closure. The effectiveness of this proposal graph SLAM is verified by simulation and experiment.