Visual SLAM and IMU fusion

Visual SLAM is easily affected by image occlusion, illumination changes, and moving objects interfere with weak texture scenes. Monocular vision cannot measure scale, and depth information cannot be directly obtained. Only a single image cannot obtain pixel depth information, which requires triangulation. method to estimate the depth of map points. In addition, the pure rotational motion of the monocular cannot be estimated, and it is easy to be lost when moving rapidly.

The fusion of vision and IMU will make up for their respective disadvantages. Visual positioning information can be used to estimate the zero offset of the IMU to reduce the divergence and accumulated errors of the IMU caused by the zero offset. IMU can provide visual positioning during fast motion, and IMU can provide scale information to avoid monocular inability to measure scale