Object identification from image data captured from a mobile aerial platforms
An image processing method includes: capturing image data from a sequence of images of a scene from different but repeatable viewpoints; associating captured images with a respective viewpoint identification; tracking a region of interest in the sequence of images; performing an eigenspace projection for the respective images to calculate respective points in the eigenspace projection; and comparing the calculated points to points previously obtained for a reference image for substantially the same viewpoints. A change between a calculated point and a corresponding previously obtained point indicates a change in the image content of the region of interest between the corresponding captured image and the corresponding reference image for a substantially same viewpoint.
1. An image processing method, comprising:
capturing image data from a sequence of images of a scene from different but repeatable viewpoints, capturing said image data with a mobile platform traveling over a repeatable path;
associating captured images with a respective viewpoint identification based on a position of the mobile platform when the captured image was captured;
tracking a region of interest in the sequence of images;
performing an eigenspace projection for respective images of the sequence to calculate respective points in the eigenspace projection; and
comparing a calculated point to a point previously obtained for a reference image for a same viewpoint;
wherein a change between a calculated point and a corresponding previously obtained point indicates a change in image content of the region of interest between a corresponding captured image and a corresponding reference image for the substantially same viewpoint.
2. A method according to claim 1 , wherein performing the eigenspace projection comprises:
using a model space created by performing Principal Component Analysis.
3. A method according to claim 2 , wherein performing Principal Component Analysis comprises:
using Singular Value Decomposition.
4. A method according to claim 1 , comprising:
a training process to create a model space for the eigenspace projection and as part of which the previously obtained points are obtained; and
a change detection process as part of which the eigenspace projections for newly captured images are performed and the comparing of the new points with the previously obtained points is performed.
5. A method according to claim 1 , wherein a change is detected by:
determining a distance in the eigenspace projection between a calculated point and a corresponding previously obtained point; and comparing the determined distance to a distance-based threshold.
6. Image processing apparatus, comprising:
apparatus for capturing image data from a sequence of images of a scene from different but repeatable viewpoints, being a mobile platform arranged to travel over a repeating path; and
one or more processors, the one or more processors being programmed to:
associate captured images with a respective viewpoint identification based on a position of the mobile platform when the captured image was captured;
track a region of interest in the sequence of images;
perform an eigenspace projection for respective images of the sequence to calculate respective points in the eigenspace projection; and
compare a calculated point to a point previously obtained for a reference image for a same viewpoint;
wherein a change between a calculated point and a corresponding previously obtained point indicates a change in image content of the region of interest between a corresponding captured image and a corresponding reference image for the substantially same viewpoint.
7. Apparatus according to claim 6 , wherein the processor for performing the eigenspace projection comprises:
means for using a model space created by performing Principal Component Analysis.
8. Apparatus according to claim 7 , wherein means for performing Principal Component Analysis comprises:
means for using Singular Value Decomposition.
9. Apparatus according to claim 6 , wherein the change is detected by:
means for determining a distance in the eigenspace projection between the calculated point and the corresponding previously obtained point, and comparing the determined distance to a distance-based threshold.
10. A computer program or plurality of computer programs on a non-transitory computer readable medium:
capturing image data from a sequence of images of a scene from different but repeatable viewpoints, from a mobile platform on a repeating path;
associating captured images with a respective viewpoint identification based on a position of the mobile platform when the image was captured;
tracking a region of interest in the sequence of images;
performing an eigenspace projection for respective images of the sequence to calculate respective points in the eigenspace projection; and
comparing a calculated point to a point previously obtained for a reference image for a same viewpoint;
wherein a change between a calculated point and a corresponding previously obtained point indicates a change in image content of the region of interest between a corresponding captured image and a corresponding reference image for the same viewpoint.