This paper presents a regularization factor based visual-inertial SLAM method considering state uncertainty in dynamic environments. To address the limitation of visual SLAM method in dynamic environments, a BA optimization method based on the regularization factor is proposed, which can reduce the impact of dynamic features on the system localization accuracy when dynamic objects account for a large number of objects by integrating the IMU measurement information and the regularization factor. Meanwhile, to deal with visual odometry accuracy degradation caused by dynamic features in dynamic environments, a visual-inertial SLAM method considering state uncertainty is proposed to reduce localization accuracy degradation caused by certain state unobservability. The method takes state uncertainty into consideration, utilizing the D-opt criterion to evaluate the uncertainty of the IMU state. The localization performance tests in VIODE datasets indicate that the effective fusion of visual-inertial information in dynamic environments is realized under limited sensor configurations and load constraints, improving the localization accuracy, and providing continuous and reliable information for perception-aware planning in active SLAM.