Posture recognition system, image correction method, and recording medium
A posture recognition system includes a sensor and a recognition apparatus, wherein the sensor captures a distance image including information on a distance to a subject, at an angular field controlled based on the distance to the subject, and the recognition apparatus includes a memory, and a processor coupled to the memory and configured to acquire the distance image and the angular field used when the sensor captures the distance image, from the sensor, and correct the acquired distance image based on a ratio between a first resolution corresponding to another angular field used when a training distance image is captured, and a second resolution corresponding to the acquired angular field, the training distance image being used for generating a learning model which recognizes a posture of the subject.
1. A posture recognition system comprising a sensor and a recognition apparatus, wherein
the sensor captures a distance image including information on a distance to a subject, at an angular field controlled based on the distance to the subject, and
the recognition apparatus includes:
a memory; and
a processor coupled to the memory and configured to:
acquire the distance image and the angular field used when the sensor captures the distance image, from the sensor,
correct the acquired distance image based on a ratio between a first resolution corresponding to another angular field used when a training distance image is captured, and a second resolution corresponding to the acquired angular field, the training distance image being used for generating a learning model which recognizes a posture of the subject,
generate a part-label image in which parts of the subject are represented by part labels, using the corrected distance image and the learning model,
recognize a skeletal structure of the subject based on the part labels in the generated part-label image, and
output skeletal structure information representing a posture of the subject based on the skeletal structure.
2. A non-transitory computer-readable recording medium storing therein an image correction program that causes a computer to execute a process, the process comprising:
acquiring a second distance image containing information on a distance from a sensor to a subject, and to be subjected to a recognition process using a learning model that is based on a first distance image, and acquiring an angular field used when the sensor captures the second distance image; and
correcting the second distance image based on a ratio between a first resolution that is based on an angular field used when the first distance image is captured, and a second resolution that is based on the angular field used when the second distance image is captured.
3. The non-transitory computer-readable recording medium according to claim 2 , wherein
the process further includes:
generating a part-label image in which parts of the subject are represented by part labels, using the corrected second distance image and the learning model,
recognizing a skeletal structure of the subject based on the part labels in the generated part-label image, and
outputting skeletal structure information indicating a posture of the subject based on the skeletal structure.
4. The non-transitory computer-readable recording medium according to claim 2 , wherein the correcting includes converting distance information in the first distance image and distance information in the second distance image into coordinate values, and correcting a value along a coordinate axis in a direction extending from the sensor to the subject in the second distance image, using the value along the coordinate axis in the direction extending from the sensor to the subject in the first distance image as a reference.
5. The non-transitory computer-readable recording medium according to claim 4 , wherein the correcting is performed in such a manner that the second resolution in one of a height direction and a width direction of the second distance image, the one being a direction in which a difference between the first resolution and the second resolution is greater, matches the first resolution in the direction in which the difference is greater.
6. The non-transitory computer-readable recording medium according to claim 4 , wherein, at the correcting, when an aspect ratio of the first resolution is different from an aspect ratio of the second resolution, the second distance image is corrected in such a manner that the aspect ratio of the second resolution matches the aspect ratio of the first resolution.
7. An image correction method executed by a processor, the image correction method comprising:
acquiring a second distance image containing information on a distance from a sensor to a subject, and to be subjected to a recognition process, using a learning model that is based on a first distance image, and acquiring an angular field used when the sensor captures the second distance image; and
correcting the second distance image based on a ratio between a first resolution that is based on an angular field used when the first distance image is captured, and a second resolution that is based on the angular field used when the second distance image is captured.
8. The image correction method according to claim 7 , further including:
generating a part-label image in which parts of the subject are represented by part labels, using the corrected second distance image and the learning model,
recognizing a skeletal structure of the subject based on the part labels in the generated part-label image, and
outputting skeletal structure information indicating a posture of the subject based on the skeletal structure.
9. The image correction method according to claim 7 , wherein the correcting includes converting distance information in the first distance image and distance information in the second distance image into coordinate values, and correcting a value along a coordinate axis in a direction extending from the sensor to the subject in the second distance image, using the value along the coordinate axis in the direction extending from the sensor to the subject in the first distance image as a reference.
10. The image correction method according to claim 9 , wherein the correcting is performed in such a manner that the second resolution in one of a height direction and a width direction of the second distance image, the one being a direction in which a difference between the first resolution and the second resolution is greater, matches the first resolution in the direction in which the difference is greater.
11. The image correction method according to claim 9 , wherein, at the correcting, when an aspect ratio of the first resolution is different from an aspect ratio of the second resolution, the second distance image is corrected in such a manner that the aspect ratio of the second resolution matches the aspect ratio of the first resolution.