We propose a novel indoor visible light positioning (VLP) system based on sensor fusion technique, which is not seen in the previous VLP works. It enhances the positioning accuracy by fusing the data collected by the built-in image sensor and motion sensors of a smartphone. The issue of effective sensor fusion is formulated as a multi-objective non-convex optimization problem. Then, we derive a closed-form optimal solution to the optimization problem and propose a low- complexity singular value decomposition based sensor fusion (SVD-SF) algorithm. Experiment results show that the proposed SVD-SF algorithm brings approximate 44% positioning accuracy improvement compared with the algorithm employing a single image sensor under the same experimental settings. Simulations are conducted to further evaluate its performance improvement under different noises conditions.