Visible light positioning (VLP) enables high-precision indoor localization, but single-sensor systems do not scale well to multi-user settings: camera-based methods suffer from motion jitter, while photodiode (PD)-based approaches are vulnerable to ambient light and multipath effects. This paper proposes a camera-photodiode fusion-based cooperative localization algorithm (C-CPD): cameras extract geometric features of a circular luminaire for coarse positioning, and PD-light-emitting diode (LED) modules measure distances to cooperative visible light communication (VLC) units via received signal strength (RSS) to refine estimates, mitigating motion distortions. We derive the Cram´er-Rao lower bound (CRLB) to characterize performance, revealing impacts of focal length, signal-to-noise ratio (SNR), and cooperative VLC units. Simulations show 0.41 cm average error, and experiments demonstrate 3.24 cm real-world error, outperforming baselines in dynamic scenarios. This work highlights multi-sensor fusion and geometric feature exploitation for robust, high-accuracy VLP, with CRLB insights guiding system design.