Classifying compartments at security checkpoints by detecting a shape of an object

1. A method for automatically detecting prohibited objects in a compartment at a security checkpoint, comprising:

receiving, by a processor of a computing system, a three-dimensional representation of a compartment from an imaging device connected to the computing system;

identifying, by the processor, a region within the compartment that has a highest likelihood of containing a prohibited object, based on a voxel classification of the three-dimensional representation using a first trained neural network, wherein the first trained neural network determines whether any voxel classifications of the three-dimensional representation correspond to a voxel classification of a prohibited material; and

classifying, by the processor, shapes of objects located within the region by feeding the shapes of the objects output by the first trained neural network into a second trained neural network to determine whether any of the shapes correspond to a shape of a prohibited object known by the second neural network, wherein the second trained neural network classifies objects located within the region as the prohibited object in response to determining that a shape output from the first trained neural network fed into the second trained neural network compares to previously classified shapes of known prohibited objects known by the second trained neural network.

2. The method of claim 1 , wherein the three-dimensional representation is a set of images in sequence of the region of the compartment, output by a computed topography (CT) scanner, further wherein each image in the set of images has a width and a height, and when stacked together, form a depth.

3. The method of claim 1 , wherein the first trained neural network is trained by:

isolating, by the processor, sub-volumes of different sizes around a voxel of a three-dimensional representation of a known prohibited material, for each voxel of the three-dimensional representation of the known prohibited material;

applying, by the processor, three-dimensional convolutions on the sub-volumes of different sizes; and

feeding, by the processor, the three-dimensional convolutions into the trained neural network so that the trained neural network learns that the convoluted sub-volumes fed into the trained neural network corresponds to a voxel classification of the known prohibited material.

4. The method of claim 1 , further comprising:

labeling, by the processor, each of the voxels determined to have a same voxel classification as a known prohibited material; and

suggesting, by the processor, the region within the compartment that has the highest likelihood of containing the prohibited object based on a location of the labeled voxels within the compartment.

5. The method of claim 4 , wherein the first trained neural network is a three-dimensional convolutional neural network (CNN) that detects a presence of the prohibited material in the compartment based on the voxel classification of the three-dimensional representation using an entire sub-volume isolated around each voxel, without requiring human inspection of an output of the imaging device.

6. The method of claim 1 , wherein the second trained neural network is trained by:

calculating, by the processor, a maximum intensity projection across each dimension of at least two orthogonal projections of a known three-dimensional representation of a prohibited object;

stacking, by the processor, each of the maximum intensity projections into a single image; and

feeding, by the processor, the single image into the second trained neural network so that the second trained neural network learns that the single image having the same maximum intensity projections corresponds to the prohibited object.

7. The method of claim 1 , wherein the second trained neural network classifies objects located within the region as a non-prohibited object in response to determining that a shape output from the first trained neural network fed into the second trained neural network does not compare to previously classified shapes of known prohibited objects known by the second trained neural network.

8. The method of claim 1 , wherein the second trained neural network is a two-dimensional convolutional neural network (CNN).

9. The method of claim 1 , wherein the second trained neural network is a three-dimensional convolutional neural network (CNN).

10. A computing system, comprising:

a processor;

a memory device coupled to the processor; and

a computer readable storage device coupled to the processor, wherein the storage device contains program executable by the processor via the memory device to implement a method for automatically detecting prohibited objects in a compartment at a security checkpoint comprising:

receiving, by a processor of a computing system, a three-dimensional representation of a compartment from an imaging device connected to the computing system;

identifying, by the processor, a region within the compartment that has a highest likelihood of containing a prohibited object, based on a voxel classification of the three-dimensional representation using a first trained neural network, wherein the first trained neural network determines whether any voxel classifications of the three-dimensional representation correspond to a voxel classification of a prohibited material; and

classifying, by the processor, shapes of objects located within the region by feeding the shapes of the objects output by the first trained neural network into a second trained neural network to determine whether any of the shapes correspond to a shape of a prohibited object known by the second neural network, wherein the second trained neural network classifies objects located within the region as the prohibited object in response to determining that a shape output from the first trained neural network fed into the second trained neural network compares to previously classified shapes of known prohibited objects known by the second trained neural network.

11. The computing system of claim 10 , wherein the first trained neural network determines whether any voxel classifications of the three-dimensional representation correspond to a voxel classification of a prohibited material.

12. The computing system of claim 10 , wherein the first trained neural network is trained by:

isolating, by the processor, sub-volumes of different sizes around a voxel of a three-dimensional representation of a known prohibited material, for each voxel of the three-dimensional representation of the known prohibited material;

applying, by the processor, three-dimensional convolutions on the sub-volumes of different sizes; and

feeding, by the processor, the three-dimensional convolutions into the trained neural network so that the trained neural network learns that the convoluted sub-volumes fed into the trained neural network corresponds to a voxel classification of the known prohibited material.

13. The computing system of claim 10 , further comprising:

labeling, by the processor, each of the voxels determined to have a same voxel classification as a known prohibited material; and

suggesting, by the processor, the region within the compartment that has the highest likelihood of containing the prohibited object based on a location of the labeled voxels within the compartment.

14. The computing system of claim 10 , wherein the first trained neural network is a three-dimensional convolutional neural network (CNN) that detects a presence of the prohibited material in the compartment based on the voxel classification of the three-dimensional representation using an entire sub-volume isolated around each voxel, without requiring human inspection of an output of the imaging device.

15. The computing system of claim 10 , wherein the second trained neural network is trained by:

calculating, by the processor, a maximum intensity projection across each dimension of at least two orthogonal projections of a known three-dimensional representation of a prohibited object;

stacking, by the processor, the maximum intensity projections into a single image; and

feeding, by the processor, the single image into the second trained neural network so that the second trained neural network learns that the single image having the same maximum intensity projections corresponds to the prohibited object.

16. A computer program product, comprising a computer readable hardware storage device storing a computer readable program code, the computer readable program code comprising an algorithm that when executed by a computer processor of a computing system implements a method for automatically detecting prohibited objects in a compartment at a security checkpoint comprising:

receiving, by a processor of a computing system, a three-dimensional representation of a compartment from an imaging device connected to the computing system;

identifying, by the processor, a region within the compartment that has a highest likelihood of containing a prohibited object, based on a voxel classification of the three-dimensional representation using a first trained neural network, wherein the first trained neural network determines whether any voxel classifications of the three-dimensional representation correspond to a voxel classification of a prohibited material; and

classifying, by the processor, shapes of objects located within the region by feeding the shapes of the objects output by the first trained neural network into a second trained neural network to determine whether any of the shapes correspond to a shape of a prohibited object known by the second neural network, wherein the second trained neural network classifies objects located within the region as the prohibited object in response to determining that a shape output from the first trained neural network fed into the second trained neural network compares to previously classified shapes of known prohibited objects known by the second trained neural network.

17. The computer program product of claim 16 , further comprising:

labeling, by the processor, each of the voxels determined to have a same voxel classification as a known prohibited material; and

suggesting, by the processor, the region within the compartment that has the highest likelihood of containing the prohibited object based on a location of the labeled voxels within the compartment.