Enhanced face detection using depth information
A method for face detection includes capturing a depth map and an image of a scene and selecting one or more locations in the image to test for presence of human faces. At each selected location, a respective face detection window is defined, having a size that is scaled according to a depth coordinate of the location that is indicated by the depth map. Apart of the image that is contained within each face detection window is processed to determine whether the face detection window contains a human face. Similar methods may also be applied in identifying other object types.
1. A method for face detection, comprising:
capturing, using a depth sensor, a depth map of a scene, the depth map comprising an array of three-dimensional coordinates;
capturing, using a camera, a two-dimensional image of the scene;
selecting, by a computer, one or more locations in the two-dimensional image to test for presence of human faces;
at each selected location, defining by the computer a respective face detection window having a size that is scaled according to an expected face size, in pixels of the image, at a depth, of the location, that is indicated by the depth map; and
processing, using the computer, a part of the two-dimensional image that is contained within each face detection window to determine whether the face detection window contains a human face.
2. The method according to claim 1 , wherein defining the respective face detection window comprises setting the size of the face detection window according to an extent, in pixels of the image, of an average face at the depth indicated by the depth map.
3. The method according to claim 1 , wherein processing the part of the image comprises normalizing the part of the image within each face detection window to a predetermined size, and applying a face detection algorithm to the normalized part of the image.
4. The method according to claim 3 , wherein applying the face detection algorithm comprises comparing the normalized part of the image within each face detection window to a single template, which is used to test for the presence of human faces at all of the selected locations.
5. The method according to claim 1 , further comprising calibrating the expected face size, using a reference image that is known to contain a face at a known depth.
6. The method according to claim 5 , wherein calibrating the expected face size comprises calibrating the expected face size by linearly scaling a size of the face in the reference image.
7. The method according to claim 1 , wherein, at each selected location, the face detection window having the size that is scaled according to the expected face size is the only face detection window defined by the computer.
8. Apparatus for face detection, comprising:
a depth sensor, which is configured to capture a depth map of a scene, the depth map comprising an array of three-dimensional coordinates;
a camera, which is configured to capture a two-dimensional image of the scene; and
a processor, which is configured to select one or more locations in the two-dimensional image to test for presence of human faces, to define, at each selected location, a respective face detection window having a size that is scaled according to an expected face size, in pixels of the image, at a depth, of the location, that is indicated by the depth map, and to process a part of the two-dimensional image that is contained within each face detection window to determine whether the face detection window contains a human face.
9. The apparatus according to claim 8 , wherein the size of the face detection window is set according to an extent, in pixels of the image, of an average face at the depth indicated by the depth map.
10. The apparatus according to claim 8 , wherein the processor is configured to normalize the part of the image within each face detection window to a predetermined size, and to apply a face detection algorithm to the normalized part of the image.
11. The apparatus according to claim 10 , wherein the face detection algorithm comprises comparing the normalized part of the image within each face detection window to a single template, which is used to test for the presence of human faces at all of the selected locations.
12. The apparatus according to claim 8 , wherein the processor is further configured to calibrate the expected face size, using a reference image that is known to contain a face at a known depth.
13. The apparatus according to claim 12 , wherein the processor is configured to calibrate the expected face size by linearly scaling a size of the face in the reference image.
14. The apparatus according to claim 8 , wherein, at each selected location, the face detection window having the size that is scaled according to the expected face size is the only face detection window the processor is configured to define.
15. A computer software product, comprising a non-transitory computer-readable medium in which program instructions are stored, which instructions, when read by a computer, cause the computer to receive a depth map, which comprises an array of three-dimensional coordinates, and to receive a two-dimensional image of the scene, to select one or more locations in the two-dimensional image to test for presence of human faces, to define, at each selected location, a respective face detection window having a size that is scaled according to an expected face size, in pixels of the image, at a depth, of the location, that is indicated by the depth map, and to process a part of the two-dimensional image that is contained within each face detection window to determine whether the face detection window contains a human face.
16. The product according to claim 15 , wherein the size of the face detection window is set according to an extent, in pixels of the image, of an average face at the depth indicated by the depth map.
17. The product according to claim 15 , wherein the instructions cause the computer to normalize the part of the image within each face detection window to a predetermined size, and to apply a face detection algorithm to the normalized part of the image.
18. The product according to claim 17 , wherein the face detection algorithm comprises comparing the normalized part of the image within each face detection window to a single template, which is used to test for the presence of human faces at all of the selected locations.
19. The product according to claim 15 , wherein the instructions further cause the computer to calibrate the expected face size, using a reference image that is known to contain a face at a known depth.
20. The product according to claim 15 , wherein, at each selected location, the face detection window having the size that is scaled according to the expected face size is the only face detection window that the instructions cause the computer to define.