US 12,243,351 B2
Gaze estimation apparatus, gaze estimation method, model generation apparatus, and model generation method
Yamato Takeuchi, Kyoto (JP); Shigenori Nagae, Kyoto (JP); Hatsumi Aoi, Kyoto (JP); and Kazuo Yamamoto, Kyoto (JP)
Assigned to OMRON CORPORATION, Kyoto (JP)
Appl. No. 17/789,234
Filed by OMRON Corporation, Kyoto (JP)
PCT Filed Jan. 10, 2020, PCT No. PCT/JP2020/000643
§ 371(c)(1), (2) Date Jun. 27, 2022,
PCT Pub. No. WO2021/140642, PCT Pub. Date Jul. 15, 2021.
Prior Publication US 2023/0036611 A1, Feb. 2, 2023
Int. Cl. G06V 40/18 (2022.01); G06V 10/82 (2022.01); G06V 10/98 (2022.01)
CPC G06V 40/193 (2022.01) [G06V 10/82 (2022.01); G06V 10/993 (2022.01)] 20 Claims
OG exemplary drawing
 
1. A gaze estimation apparatus comprising a processor configured with a program to perform operations comprising:
obtaining calibration information comprising feature information and true value information, the feature information being about a gaze of an eye of a target person looking in a predetermined direction, the true value information indicating a true value for the predetermined direction in which the eye of the target person is looking;
obtaining a target image comprising the eye of the target person;
estimating a gaze direction of the target person comprised in the target image using a learned estimation model generated through machine learning, the learned estimation model being trained through the machine learning to output, in response to an input of calibration information for learning and a target image for learning obtained from a subject, an output value fitting answer information indicating a true value of a gaze direction of the subject comprised in the target image for learning, estimating the gaze direction comprises inputting the obtained target image and the obtained calibration information into the learned estimation model and performing a computational operation of the learned estimation model to obtain, from the learned estimation model, an output value corresponding to a result from estimating the gaze direction of the target person comprised in the target image; and
outputting information about the result from estimating the gaze direction of the target person, wherein
the learned estimation model is trained by obtaining a plurality of learning datasets each comprising a target image for learning comprising the eye of the subject and answer information indicating the true value for the gaze direction of the subject comprised in the respective target image for learning, and for each dataset of the plurality of learning datasets, performing the machine learning of the estimation model to output the output value fitting a corresponding piece of the answer information.