Aiming at the problem that single visual sensor SLAM technology has low accuracy and poor reliability in dynamic environment, which leads to the inability to accurately estimate the camera pose, a visual SLAM algorithm based on weighted static feature points and selective optimization (CW-SLAM) was proposed. Firstly, dynamic feature point detection was added to the front end. After using the GC-RANSAC algorithm to separate the inner/outer points and fit the optimal basic matrix, a weighted static feature detection method was designed to eliminate the error constraints from the dynamic features, improve the matching accuracy, and use the stable features for back-end pose optimization. Secondly, the factor graph model was used to construct a new structure with vision as the main system and IMU as the auxiliary system. By introducing the auxiliary system IMU odometer factor to constrain the main system error, and receiving the VIO factor to realize the motion prediction and pose optimization. Finally, a selective optimization strategy was proposed to eliminate the influence of temporary static targets. After clustering the loopback key frames, the selective optimization of the constraint group was established according to the factor graph optimization model to filter out the false positive loopback hypothesis. Compared with the classical SLAM algorithm, the effectiveness of the algorithm was verified on the TUM public dataset and in the real environment. The experimental results showed that the algorithm can effectively suppress the influence of dynamic and temporary stationary targets on pose estimation, and improve the accuracy and reliability.