In order to improve the positioning performance of AGV (automated guided vehicle) in complex visual environment and reduce the hardware cost, VIO (visual inertia odometer) based on AprilTag and fuzzy complementary filter was proposed. EKF (extended Kalman filter) was used to integrate the measurement data of gyroscope, magnetometer and encoder to calculate the course angle, which is used for dead reckoning. Fuzzy calculations were used to determine tag weight based on recognition distance and motion speed of camera. The weighted sum of AprilTags was used for visual positioning to reduce multi-tag positioning error. The complementary fusion coefficients were derived from the mean value of tag weight. Finally, the visual positioning and the dead reckoning results were fused by complementary coefficients to improve the VIO positioning accuracy. Experimental results show that the proposed VIO implemented on AGV has a 41.84 mm positioning accuracy, which is 52.20% and 20.75% higher than that of dead-reckoning and AprilTag-VIO with traditional Kalman filter.