Systems and methods for gaze-tracking
Systems and methods are provided for predicting an eye gaze location of an operator of a computing device. In particular, the method generates an image grid that includes regions of interest based on a facial image. The facial image is based on a received image frame of a video stream that captures the operator using the computing device. The image grid further includes a region that indicate rotation information of the face. The method further uses a combination of trained neural networks to extract features of the regions of interest in the image grid and predict the eye gaze location on the screen of the computing device. The trained set of neural networks includes a convolutional neural network. The method optionally generate head pose pitch, roll, and yaw information to improve accuracy of predicting the location of an eye gaze.
1. A computer-implemented method for predicting a location of an eye gaze of an operator, the method comprising:
receiving an input image, wherein the input image depicting the operator gazing at a location on a screen of a device;
extracting a plurality of regions from the input image, wherein the plurality of regions include:
one or more eyes,
a face, and
a region indicating at least a position and a rotation of the face;
generating, based on the extracted plurality of regions, an image grid, wherein the image grid includes a plurality of image data, each image data corresponding to a distinct region of the plurality of regions;
determining, based on a combination of a convolutional network and a fully-connected network using the image grid, a predicted location of the eye gaze, wherein the predicted location includes a two-dimensional location of the eye gaze on the screen of the device; and
causing, based on the predicted location, one or more actions associated with an operation of the device.
2. The computer-implemented method according to claim 1 , the method further comprising:
generating, based on the input image, head pose information of the operator, wherein the head pose information includes a direction of a head of the operator relative to the screen of the device;
generating, based on the image grid using the convolution network trained with a gaze-data prediction model, a multi-dimensional vector; and
generating, based on a combination of the multi-dimensional vector and the head pose information using the fully-connected network, the predicted location of the eye gaze.
3. The computer-implemented method according to claim 1 , the method further comprising:
transmitting the predicted location of the eye gaze to an output controller, causing an update to a gaze location indicator on the screen of the device.
4. The computer-implemented method according to claim 1 , the method further comprising:
receiving an input image from an RGB camera, wherein the input image is based on an RGB color space.
5. The computer-implemented method according to claim 1 , wherein the image grid includes a combination of:
the one or more eyes,
the face,
a nose,
one or more ears, and
the region indicating at least a position and a rotation of the face.
6. The computer-implemented method according to claim 1 , wherein the convolutional neural network predicts one or more features of the one or more eyes based on one or more of:
a feature of the face, and
a feature of the region indicating at least a position and a rotation of the face in the image grid.
7. The computer-implemented method according to claim 1 , wherein the head pose information includes one or more of a two-dimensional coordinates and a three-dimensional locations of one or more of:
a corner of an eye,
a tip of a nose,
a corner of a mouth, or
a tip of a chin.
8. A system for predicting a location of an eye gaze of an operator, the system comprising:
a processor; and
a memory storing computer-executable instructions that when executed by the processor cause the system to:
receive an input image, wherein the input image depicting the operator gazing at a location on a screen of a device;
extract a plurality of regions from the input image, wherein the plurality of regions include:
one or more eyes,
a face, and
a region indicating at least a position and a rotation of the face;
generate, based on the extracted plurality of regions, an image grid, wherein the image grid includes a plurality of image data, each image data corresponding to a distinct region of the plurality of regions;
determine, based on a combination of a convolutional network and a fully-connected network using the image grid, a predicted location of the eye gaze, wherein the predicted location includes a two-dimensional location of the eye gaze on the screen of the device; and
cause, based on the predicted location, one or more actions associated with an operation of the device.
9. The system according to claim 8 , the computer-executable instructions that when executed by the processor further cause the system to:
generate, based on the input image, head pose information of the operator, wherein the head pose information includes a direction of a head of the operator relative to the screen of the device;
generate, based on the image grid using the convolution network trained with a gaze-data prediction model, a multi-dimensional vector; and
generate, based on a combination of the multi-dimensional vector and the head pose information using the fully-connected network, the predicted location of the eye gaze.
10. The system according to claim 8 , the computer-executable instructions that when executed by the processor further cause the system to:
transmit the predicted location of the eye gaze to an output controller, causing an update to a gaze location indicator on the screen of the device.
11. The system according to claim 8 , the computer-executable instructions that when executed by the processor further cause the system to:
receive an input image from an RGB camera, wherein the input image is based on an RGB color space.
12. The system according to claim 8 , wherein the image grid includes a combination of:
the one or more eyes,
the face,
a nose,
one or more ears, and
the region indicating at least a position and a rotation of the face.
13. The system according to claim 8 , wherein the convolutional neural network predicts one or more features of the one or more eyes based on one or more of:
a feature of the face, and
a feature of the region indicating at least a position and a rotation of the face in the image grid.
14. The system according to claim 8 , wherein the head pose information includes one or more of a two-dimensional coordinates and a three-dimensional locations of one or more of:
a corner of an eye,
a tip of a nose,
a corner of a mouth, or
a tip of a chin.
15. A computer-readable storage medium storing computer-executable instructions that when executed by a processor cause a computer system to:
receive an input image, wherein the input image depicting the operator gazing at a location on a screen of a device;
extract a plurality of regions from the input image, wherein the plurality of regions include:
one or more eyes,
a face, and
a region indicating at least a position and a rotation of the face;
generate, based on the extracted plurality of regions, an image grid, wherein the image grid includes a plurality of image data, each image data corresponding to a distinct region of the plurality of regions;
determine, based on a combination of a convolutional network and a fully-connected network using the image grid, a predicted location of the eye gaze, wherein the predicted location includes a two-dimensional location of the eye gaze on the screen of the device; and
cause, based on the predicted location, one or more actions associated with an operation of the device.
16. The computer-readable storage medium of claim 15 , the computer-executable instructions that when executed by the processor further cause the system to:
generate, based on the input image, head pose information of the operator, wherein the head pose information includes a direction of a head of the operator relative to the screen of the device;
generate, based on the image grid using the convolution network trained with a gaze-data prediction model, a multi-dimensional vector; and
generate, based on a combination of the multi-dimensional vector and the head pose information using the fully-connected network, the predicted location of the eye gaze.
17. The computer-readable storage medium of claim 15 , the computer-executable instructions that when executed by the processor further cause the system to:
transmit the predicted location of the eye gaze to an output controller, causing an update to a gaze location indicator on the screen of the device.
18. The computer-readable storage medium of claim 15 , the computer-executable instructions that when executed by the processor further cause the system to:
receive an input image from an RGB camera, wherein the input image is based on an RGB color space.
19. The computer-readable storage medium of claim 15 ,
wherein the image grid includes a combination of:
the one or more eyes,
the face,
a nose,
one or more ears, and
the region indicating at least a position and a rotation of the face, and
wherein the convolutional neural network predicts one or more features of the one or more eyes based on one or more of:
a feature of the face, and
a feature of the region indicating at least a position and a rotation of the face in the image grid.
20. The computer-readable storage medium of claim 15 , wherein the head pose information includes one or more of a two-dimensional coordinates and a three-dimensional locations of one or more of:
a corner of an eye,
a tip of a nose,
a corner of a mouth, or
a tip of a chin.