A portable real-time scene recognition system on smartphone

In the context of mapping and mobile navigation services, the recognition of the urban infrastructure plays an important role. However, majority of the existing recognition methods cannot be effectively run on the smartphones due to their large computational costs and high storages. In this paper, a fast scene recognition approach is introduced by fusing the outputs of inertial sensors and cameras of smartphones for real-time applications in large scenes. The proposed algorithm possesses the virtues of implementation easiness, storage efficiency and computation significance. Experimental results on the datasets of outdoor scenes and UKBench datasets demonstrate the effectiveness and robustness of the proposed algorithm.