Method for determining bias in an inertial measurement unit of an image acquisition device
View Patent ↗A method for determining bias in an inertial measurement unit of an image acquisition device comprises mapping at least one reference point within an image frame into a 3D spherical space based on a lens projection model for the image acquisition device to provide a respective anchor point in 3D space for each reference point. For reference points within a given image frame, an estimate of frame-to-frame motion at the reference point between the given frame and a previously acquired frame is obtained; a measure of device orientation for an acquisition time of the reference point in the given frame and the previously acquired frame, the measure including a bias component, is obtained from the inertial measurement unit; a corresponding anchor point is projected in 3D space according to a difference in the measure of device orientation in the given frame and the previously acquired frame to provide a 3D vector V m ; a result of the estimated frame-to-frame motion for the point from the given frame is projected into the previously acquired frame into 3D space to provide a 3D vector V e ; and a cross product V c of the 3D vectors V m and V e is used to update a bias component value.
1. A method for determining bias in an inertial measurement unit of an image acquisition device comprising:
mapping at least one reference point within an image frame into a 3D spherical space based on a lens projection model for the image acquisition device to provide a respective anchor point in 3D space for each reference point;
for at least one of said at least one reference points within a given image frame:
obtaining an estimate of frame-to-frame motion at said reference point between said given frame and a previously acquired frame;
obtaining from said inertial measurement unit a measure of device orientation for an acquisition time of said reference point in said given frame and said previously acquired frame, said measure including a bias component;
projecting a corresponding anchor point in 3D space according to a difference in said measure of device orientation in said given frame and said previously acquired frame to provide a 3D vector V m ;
projecting a result of said estimated frame-to-frame motion for said point from said given frame into said previously acquired frame into 3D space to provide a 3D vector V e ;
providing a cross product V c of said 3D vectors V m and V e , and
obtaining an updated bias component value as a function of said cross product V c for a plurality of reference points in one or more image frames.
2. A method according to claim 1 comprising performing said mapping only once for a given lens projection model.
3. A method according to claim 1 comprising acquiring each frame with a rolling shutter so that the acquisition time for one reference point within a given frame is different than for a reference point from another row of said frame.
4. A method according to claim 1 wherein said previously acquired frame is an immediately previously acquired frame.
5. A method according to claim 1 comprising determining device orientation Q as:
Q=Q ( t−dt )+ {dot over (Q)}dt
where
Q is a quaternion function Q(t) representing device orientation at time t,
dt is a sample interval,
{dot over (Q)} is a quaternion update function: {dot over (Q)}=0.5 Q(t−dt)·[0, ω−B],
ω=[ω x , ω y , ω z ] is an instantaneous device angular velocity; and
B is said bias component.
6. A method according to claim 5 comprising normalising said device orientation Q to maintain said orientation on a unit sphere in said 3D spherical space.
7. A method according to claim 5 further comprising calculating said vector V m as follows: V m =V·(Q(t) −1 ·Q(t−df)), where V is an anchor point in 3D space, t is said acquisition time for reference point in said given frame; and df is an interval between said given frame and said previously acquired frame.
8. A method according to claim 1 wherein obtaining said updated bias component value is determined as follows: B=B −1 +ηsign(V c ), where η<<1 and η>0 and where B −1 is a previous estimate of said bias component.
9. A method according to claim 1 wherein said reference points are distributed in a regular pattern across an image.
10. A method according to claim 1 further comprising selecting only reference points for a given image frame which are determined to be valid.
11. A method according to claim 1 further comprising selecting only reference points for a given image frame for which some motion is determined.
12. An image processing device arranged to perform the method of claim 1 .
13. A computer program product comprising computer readable instructions, which when executed in an image processing device are arranged to perform the method of claim 1 .