Visual, inertial, and ultra-wideband (UWB) are the most commonly used sensors in indoor positioning scenarios. Visual sensors can capture environmental images and extract texture information from the scene, enabling the creation of an environment map while performing positioning. However, visual positioning technology has a relatively low precision, and visual sensors cannot function in strong or low light conditions. Inertial sensors have a high signal collection frequency and do not fail, providing high dynamic and accurate positioning within a short period of time. However, the positioning precision is limited by the drift of the sensors, and the positioning precision will decrease significantly over a long period of time. UWB positioning technology has a relatively high positioning precision and does not have the problem of cumulative error. It can perform positioning in a fixed global coordinate system. However, UWB positioning technology is susceptible to Non-Line-Of-Sight (NLOS) errors. If the three sensors are effectively and reasonably fused, the positioning accuracy and adaptability to indoor complex environments can be effectively improved.For this purpose, a graph-optimization-based Visual/inertial/UWB Fusion Positioning Algorithm (VIUFPA) is proposed. Firstly, visual inertial odometry based on point-line features is used to estimate the local pose, and the point-line feature extraction improves the positioning accuracy and robustness of the visual positioning system in scenes with changing lighting, weak textures, and fast camera movement. Secondly, the Robust Kalman Filtering (RKF) is designed to preprocess the UWB distance measurement values, eliminate the NLOS errors and abnormal distance measurement values, and then a UWB positioning algorithm based on RKF is constructed to provide global positioning information. Then, the UWB positioning algorithm output information is fused with the visual inertial odometry output positioning information using graph optimization to achieve high-precision indoor positioning.Finally, the validity of the proposed algorithm was verified through simulation experiments and real indoor scene experiments. First, the NLOS error suppression performance of the RKF algorithm in the presence of pedestrian interference was analyzed; second, a simulation experiment was conducted using the EuRoC dataset, where the UWB distance measurement values were simulated based on the information provided by the dataset, proving that the VIUFPA algorithm has high precision and robustness in complex environments; finally, an experimental platform was set up to conduct positioning experiments in an office with normal lighting conditions and an underground parking with weak lighting conditions, and the experimental results showed that, the proposed algorithm achieves an average positioning accuracy improvement of about 72.09% compared with the visual inertial odometry in low light, weak textures, or obstacle occlusion environments, and an average positioning accuracy improvement of about 46.15% compared with pure UWB positioning algorithm. The proposed algorithm can provide higher precision and stronger robustness positioning results in indoor environments.