Method and system for fusing data arising from image sensors and from motion or position sensors
AssigneeCommissariat A L'energie Atomique Et Aux Energies Alternatives
InventorsJulien Michot, Adrien Bartoli, François Gaspard
Life of this patent
Estimating the term…
Abstract
A method and a system are provided integrating inter-pose constraints, which are weighted in a step of “bundle adjustment” in a localization application (SLAM). The inertial data associated with a motion model serve to predict the pose of the system at the instant of a new image. This prediction is then used to construct an inter-pose constraint. A technique arising from learning theory selects the weighting coefficient (ratio of the noise of the covariance signals) in an efficient manner. The inter-pose constraint thus weighted is then integrated within the cost function of the bundle adjustment.
Claims8
A method for fusing data arising from a first image sensor and from a motion and/or position measuring module, the said first sensor and the said motion and/or position measuring module being integral and forming a device, the method being executed by a processing module in order to localize the said device, comprising:
- a first step consisting of acquiring several points on the images in order to define a current pose ptc of the device, including image camera and motion and/or position measuring module, for a given instant t,
- a prediction step of the predicted pose {circumflex over (p)}tk of the device (camera and motion/position sensor) for the same given instant t and using the measurements from the motion and/or position measuring module,
- a step where we determine an associated constraint εk for the sensor k of the device (camera c and sensor k),
- a step where we automatically determine a weighting coefficient λt of the said constraint,
- an estimation step of the current pose pt of the said device,
- a step of simultaneous optimization of the poses of the device (camera+sensor) by the bundle adjustment technique using the following cost function:
- E
- BA
-
- (
- X
- t
- ,
- Z
- t
- )
- =
- ∑
- t
- ′
- =
- t
- -
- (
- N
- -
- 1
- )
- t
-
- ɛ
- c
- 2
-
- (
- [
- p
- t
- ′
-
- s
- ]
- ,
- z
- t
- ′
- c
- )
- +
- ∑
- t
- ′
- =
- t
- -
- (
- N
- -
- 1
- )
- t
-
- E
- p
-
- (
- [
- p
- t
- ′
-
- s
- ]
- ,
- Z
- t
- )
- with
- Ep(xt,Zt)=εc2(xt,ztc)+λt2εk2(xt,ztk)
- εc2:
- reprojection errors of the scene in the images
- xt:
- =[ps] the pose of the device and the scene
- ztc:
- =the observations of the scene detected in the image of the camera c at the instant t
- Zt=[ztcztk] combination of all the measurements for all the camera c sensors and of the sensors k of motion and/or position module at an instant t.
Dependencies are inferred from the claim text.