Video surveillance system based on larger pose face frontalization
A video surveillance system is provided. The system includes a device configured to capture an input image of a subject located in an area. The system further includes a processor. The processor estimates, using a three-dimensional Morphable Model (3DMM) conditioned Generative Adversarial Network, 3DMM coefficients for the subject of the input image. The subject varies from an ideal front pose. The processor produces, using an image generator, a synthetic frontal face image of the subject of the input image based on the input image and coefficients. An area spanning the frontal face of the subject is made larger in the synthetic than in the input image. The processor provides, using a discriminator, a decision of whether the subject of the synthetic image is an actual person. The processor provides, using a face recognition engine, an identity of the subject in the input image based on the synthetic and input images.
1. A video surveillance system, comprising:
an image capture device configured to capture an input image of a subject located in a target area;
a processor, configured to
estimate, using a three-dimensional Morphable Model (3DMM) conditioned Generative Adversarial Network (GAN), 3DMM coefficients for the subject of the input image, wherein the subject varies from an ideal front pose;
produce, using an image generator, a synthetic frontal face image of the subject of the input image based on the input image and the 3DMM coefficients, wherein an area spanning the frontal face of the subject is made larger in the synthetic frontal face image than in the input image;
provide, using a discriminator, a decision indicative of whether the subject of the synthetic frontal face image is an actual person; and
provide, using a face recognition engine, an identity of the subject in the input image based on the synthetic frontal face image and the input image.
2. The video surveillance system of claim 1 , wherein the face recognition engine regularizes the identity of the subject in the input image relative to the synthetic frontal face image.
3. The video surveillance system of claim 1 , wherein the discriminator provides the decision by distinguishing the synthetic frontal face image from at least one ground truth frontal face image.
4. The video surveillance system of claim 1 , wherein the discriminator is pre-trained to minimize a classification loss between the input image and the synthetic frontal face image.
5. The video surveillance system of claim 1 , wherein the discriminator comprises a linear layer that generates a two-dimensional vector, with each of the two dimensions representing a respective probability of the subject of the synthetic frontal face image being the actual person.
6. The video surveillance system of claim 1 , wherein the image generator is configured to mislead the discriminator to classify synthetic frontal face images to be real images using an objective function in order to improve both generator performance and face recognition engine performance.
7. The video surveillance system of claim 1 , wherein the face recognition engine is pre-trained to maximize a probability of the subject of the input image being classified as a one of a plurality of ground truth identities.
8. The video surveillance system of claim 1 , wherein the generator is guided to generate the synthesized frontal face image so as to include an identity-preserved frontal face relative to the subject of the input image.
9. The video surveillance system of claim 1 , wherein the image generator generates the synthetic frontal face image using a symmetry loss to account for symmetry variations between opposing face sides.
10. The video surveillance system of claim 1 , wherein the 3DMM is configured to define a three-dimensional face shape and texture in a principal component analysis space.
11. The video surveillance system of claim 1 , wherein the input image is used by the 3DMM to compensate for a loss of discriminative identity features of the subject.
12. The video surveillance system of claim 1 , wherein the 3DMM conditioned GAN uses a 3DMM fitting process to estimate the 3DMM coefficients for the input image.
13. The video surveillance system of claim 1 , wherein the processor is further configured to actuate a lock to secure the subject from the target area or a region within the target area, responsive to the identity of the subject.
14. The video surveillance system of claim 1 , wherein the processor is further configured to actuate a lock to release the subject from the target area or a region within the target area, responsive to the identity of the subject.
15. The video surveillance system of claim 1 , wherein the processor is further configured to log a detection of the subject as posing a danger and generate a remote alarm indicative of the detection, responsive to the identity of the subject.
16. A computer-implemented method for video surveillance, comprising:
capturing, by an image capture device, an input image of a subject located in a target area;
estimating, by a processor using a three-dimensional Morphable Model (3DMM) conditioned Generative Adversarial Network (GAN), 3DMM coefficients for the subject of the input image, wherein the subject varies from an ideal front pose;
producing, by the processor using an image generator, a synthetic frontal face image of the subject of the input image based on the input image and the 3DMM coefficients, wherein an area spanning the frontal face of the subject is made larger in the synthetic frontal face image than in the input image;
providing, by the processor using a discriminator, a decision indicative of whether the subject of the synthetic frontal face image is an actual person; and
providing, by the processor using a face recognition engine, an identity of the subject in the input image based on the synthetic frontal face image and the input image.
17. The computer-implemented method of claim 16 , wherein the face recognition engine regularizes the identity of the subject in the input image relative to the synthetic frontal face image.
18. The computer-implemented method of claim 16 , wherein the discriminator provides the decision by distinguishing the synthetic frontal face image from at least one ground truth frontal face image.
19. The computer-implemented method of claim 16 , wherein the discriminator is pre-trained to minimize a classification loss between the input image and the synthetic frontal face image.
20. A computer program product for video surveillance, the computer program product comprising a non-transitory computer readable storage medium having program instructions embodied therewith, the program instructions executable by a computer to cause the computer to perform a method comprising:
capturing, by an image capture device, an input image of a subject located in a target area;
estimating, by a processor using a three-dimensional Morphable Model (3DMM) conditioned Generative Adversarial Network (GAN), 3DMM coefficients for the subject of the input image, wherein the subject varies from an ideal front pose;
producing, by the processor using an image generator, a synthetic frontal face image of the subject of the input image based on the input image and the 3DMM coefficients, wherein an area spanning the frontal face of the subject is made larger in the synthetic frontal face image than in the input image;
providing, by the processor using a discriminator, a decision indicative of whether the subject of the synthetic frontal face image is an actual person; and
providing, by the processor using a face recognition engine, an identity of the subject in the input image based on the synthetic frontal face image and the input image.