Monocular visual odometers based on deep learning have achieved excellent performance. However, their reliability tends to deteriorate in the presence of dynamic objects. Optical flow describes the displacement vector of each pixel from one frame to the next in an image sequence. While it can be used to detect dynamic objects in a scene, it performs poorly in some situations, particularly during turns where it masks much of the image. At high turn speeds, it may even mask the entire image. Therefore, this method is unsuitable for dynamic object detection in visual odometry (VO) tasks. To address this issue, this paper introduces a fast visual odometry method named MoMaskVO, capable of suppressing dynamic interference effectively. Our method can capture occlusion information resulting from the motion of dynamic objects in the scene through forward and backward consistency detection of the optical flow. This approach can also remove the dynamic optical flow caused by moving objects directly from the optical flow image. By doing so, we improve the accuracy and reliability of visual odometry.