Is pose estimation using images necessary in visual-inertial SLAM

Question

In visual SLAM, pose estimation is performed between two adjacent frames or keyframes using feature detection and pose estitmation from corresponding features.

However in visual-inertial SLAM, pose information can be calculated from acceleroemeter and gyro. Is visual information still used to estimation poses? I have read several papers about SLAM, but authors don't mention this explicitly.

Edit:

From the link:

Full 3D Pose (Scale, Rotation and Translation) Estimation using Gyro and Acceleromter sensors fusion

Pose estimation from IMU sensor is incorrect due to bias. Then IMU sensor will not help in typical situation, but only in the situation where few textures are found and feature matching won't work?

Answer 1

No, nothing is "necessary". You can estimate the pose of the robot perfectly legitimately using only IMU data. You can also estimate it perfectly legitimately using only image data. But it won't be that good, there will be errors in any sensor.

Localization is not a decision problem. It can be tempting to "switch" between sensors and think really hard about when one sensor is better than another. The community has determined that is the wrong way to do business. It's much better (the story goes) to use all sensors merged together. The community has agreed since the days of Apollo that an average of all position estimates, weighted by uncertainty, is best.

Almost any practitioner will tell you the best way to localize a robot using IMU and vision is V+INS SLAM, of which there are several open-source implementations available. Currently ORB-SLAM or MAPLAB is very popular (I'm told).

You may also want to peruse this q/a post or (disclaimer, I said a similar thing there).

Answer 2

The typical way of doing visual-inertial SLAM is to create a short accumulated trajectory from IMU and correcting it by visual observation. The pose estimation by the images wouldn't be performed but implicitly affect during the joint optimization.