Monocular SLAM with video, gyro and accelerometer

I am trying to get cameramovement and orientation. My possible input values are video, gyro and accelerometer.

I read a lot about the problems of monocular SLAM, I am wondering if the problem becomes easier with the help of accelerometer and gyro. At least the gyro by itself seems to return pretty solid data for camera orientation, but the accelerometer is no where near as accurate as it needs to be to determine camera movement, so I was hoping a cloud of tracking points could help here.

It’s also worth pointing out that this does NOT have to be a real time system. If this crunches data for 10 minutes for a 30 second clip, I don’t really care.

Any pointers on functions that OpenCV provides that could help with this problem or an outline of an algorithm that can solve this?


sounds like a good idea, unfortunately, opencv has nothing for that builtin