With the rapid development of the Internet of Things, smart devices require high-precision, real-time location services. However, only two-dimensional positioning can be achieved in existing visible light positioning technologies; furthermore, the system is complex and has a low positioning accuracy and extended positioning time. To address these issues, this study proposes a visible light three-dimensional positioning system based on two light-emitting diodes (LEDs) and image sensors. First, a simulation model is established by combining the pinhole imaging model and the distortion model. Next, a stripe search algorithm based on bisection and a double pointer is proposed to rapidly determine the position of the LED on the image. Finally, the geometric features of the LEDs on the image are used to achieve three-dimensional positioning. After calculations, in a 100 cm×100 cm×300 cm space, the average positioning error of the simulation experiment is $\mathrm{0.79}\mathrm{c}\mathrm{m}$, and the average positioning error of the physical experiment is $\mathrm{5.61}\mathrm{c}\mathrm{m}$. The average positioning time is $\mathrm{64.13}\mathrm{m}\mathrm{s}$. In addition, the time performance of the proposed stripe search algorithm is improved by $\mathrm{70.06}\mathrm{\%}$ compared with that of the linear stripe search algorithm. Therefore, the proposed system can provide a good three-dimensional positioning service.