To make the operation of visible indoor positioning simpler and independent of line-of-sight view and position registration, photodiodes are used as the receivers, and the system is optimized in the hardware by assembling compatible photodiodes and compact LED. In the algorithm part, based on fingerprint technology, the Gauss Process Regression (GPR) algorithm is used for location. Firstly, the GPR environment is modeled and a strength distribution model is constructed. Then, Bayesian theory is used to obtain the expression of location. The experiment is carried out in a 3m×3 m platform area and 25 targets are tested. Compared with similar indoor positioning systems, the proposed system has higher positioning accuracy and achieves centimeter-level positioning. The proposed system can accurately locate moving targets for a long time and has good robustness.