Object detection system

1. An object detection system comprising:

an image sensor configured to obtain input image data of a body of water;

a processor for executing modules;

a memory containing the modules, the executed modules including:

an image slice extraction module configured to extract a sequence of 2-D slices from the input image data;

a 3-D analysis module configured to perform a 3-D connected region analysis on the sequence of 2-D slices;

an invariant feature computation module configured to compute image 3-D invariant features in the input image data;

an image filter module configured to perform coarse filtering based on the image 3-D invariant features, the coarse filtering including comparing the image 3-D invariant features in the input image data to a data store of trained 3-D invariant features associated with known objects; and

an image recognition module configured to perform fine recognition to identify at least one of the known objects from the data store within the input image data.

2. The system of claim 1 , further comprising:

an object training module configured to extract training 3-D invariant features of the known objects in training data, and store the training 3-D invariant features as the trained 3-D invariant features and the known objects in the data store.

3. The system of claim 1 , wherein performing the fine recognition includes applying a metric including a Hausdorff metric.

4. The system of claim 1 , wherein the image slice extraction module is further configured to apply diffusion equations to generate the sequence of 2-D slices.

5. The system of claim 1 , wherein the image 3-D invariant features include moment invariants.

6. The system of claim 1 , wherein the image sensor is located on an aircraft.

7. The system of claim 6 , wherein the aircraft is a helicopter, fixed wing aircraft, or unmanned aerial vehicle.

8. A computerized method for detecting objects, comprising:

obtaining, at a processor, input image data of a body of water from an image sensor;

extracting, using the processor, a sequence of 2-D slices from the input image data;

performing a 3-D connected region analysis, using the processor, on the sequence of 2-D slices;

determining, using the processor, image 3-D invariant features in the input image data;

performing, using the processor, coarse filtering based on the image 3-D invariant features, the coarse filtering including comparing the image 3-D invariant features in the input image data to a data store of trained 3-D invariant features associated with known objects; and

performing fine recognition to identify at least one of the known objects from the data store within the input image data.

9. The method of claim 8 , further comprising:

training the data store, the training comprising

extracting training 3-D invariant features of the known objects in training data; and

storing the training 3-D invariant features as the trained 3-D invariant features and the known objects in the data store.

10. The method of claim 8 , wherein performing the fine recognition includes applying a metric including a Hausdorff metric.

11. The method of claim 8 , wherein the extracting includes using diffusion equations to generate the sequence of 2-D slices.

12. The method of claim 8 , wherein the image 3-D invariant features include moment invariants.

13. The method of claim 8 , wherein performing the 3-D connected region analysis includes analyzing, at the processor, volumetric pixel elements.

14. A nontransitory computer readable medium having stored thereon instructions that, when executed by a processor, cause the processor to perform a series of operations to detect objects, the detecting including:

obtaining, at a processor, input image data of a body of water from an image sensor;

extracting, using the processor, a sequence of 2-D slices from the input image data;

performing a 3-D connected region analysis, using the processor, on the sequence of 2-D slices;

determining, using the processor, image 3-D invariant features of at least one object in the input image data;

performing, using the processor, coarse filtering based on the image 3-D invariant features, the coarse filtering including comparing the image 3-D invariant features of the at least one object to a data store of trained 3-D invariant features associated with known objects; and

performing fine recognition to determine if the at least one object in the input image data corresponds to one of the known objects.

15. The nontransitory computer readable medium of claim 14 , wherein the detecting further includes:

training the data store, the training comprising:

extracting training 3-D invariant features of the known objects in training data; and

storing the training 3-D invariant features as the trained 3-D invariant features and the known objects in the data store.

16. The nontransitory computer readable medium of claim 14 , wherein performing the fine recognition includes applying a metric including a Hausdorff metric.

17. The nontransitory computer readable medium of claim 14 , wherein the extracting includes using diffusion equations to generate the sequence of 2-D slices.

18. The nontransitory computer readable medium of claim 14 , wherein the image 3-D invariant features include moment invariants.

19. The nontransitory computer readable medium of claim 14 , wherein performing the 3-D connected region analysis includes analyzing, at the processor, volumetric pixel elements.

20. The nontransitory computer readable medium of claim 14 , wherein the extracting includes extracting different size objects through a series of filtering operations.