In recent years, a visible light positioning (VLP) technology based on complementary metal-oxide-semiconductor sensors has been widely studied due to its high precision and high robustness. However, the existing VLP algorithm based on image sensors often fails to achieve a good positioning effect when the camera is tilted. In order to solve this problem, we propose a neural network algorithm to correct the error caused by the tilt angle of the camera. Because when the tilt angle is different, the LED image captured by the camera will be different and produce different characteristics. By extracting these features and using neural networks to establish the relationship between the characteristics of the LED image and the distance between the receiving and sending terminal, we finally achieve the positioning of the camera by the triangulation algorithm. Experiments demonstrate that our positioning algorithm can achieve high-precision positioning and can be applied to most indoor positioning systems.