Optical flow inputs

Question

I'm reading how the optical flow of my quadcopter works. The output of the optical flow that interests me is the linear (horizontal) velocity: $v_x$ and $v_y$. Here (page 10) it says that it fuses the IMU information with the camera images to do that, and the algorithm used is Optical Flow.

When I investigate about optical flow, it says that the algorithm uses the angular velocity from the IMU and explains how to fuse it with the camera. But the angular velocity is given by the gyroscope, so why do I need an accelerometer?

Do you know where the fusion between gyroscope, accelerometer to obtain the angular velocity (or whatever the Optical Flow algorithm uses) is better explained?

Answer 1

You are correct, you do not really need an accelerometer for computing optical flow - just the angular velocities along with an altitude estimate will be sufficient for computing velocities in global, metric units. For example, the commercially available PX4Flow optical camera does not contain an accelerometer, only a gyroscope.

But in the case of the slides you linked to, it looks like they are estimating roll and pitch angles both from the gyroscope as well as from the accelerometer just for redundancy purposes (slides 24, 25).