Detection device, learning device, detection method, learning method, and recording medium
View Patent ↗A detection device includes an image segmenter and a detector. The image segmenter cuts out a first region image and a second region image from an image of a vehicle interior that is acquired from an imaging device. The first region image shows at least a portion of a first part of a body of an occupant. The second region image shows at least a portion of a region of the vehicle interior around the first part or at least a portion of a second part of the body of the occupant. The detector detects an orientation of the first part of the body of the occupant based on a feature amount of the first region image and a feature amount of the second region image.
1. A detection device comprising:
an image segmenter configured to cut out, from one image of a vehicle interior acquired from an imaging device, a first region image showing at least a portion of a first part of a body of an occupant, and a second region image from said one image showing at least a portion of a region of the vehicle interior around the first part; and
a detector configured to detect an orientation of the first part of the body of the occupant based on a feature amount of the first region image and a feature amount of the second region image,
wherein the first part is a face or a head of the occupant, and the feature amount of the first region image is a positional relationship between facial landmarks of the occupant,
wherein the region of the vehicle interior is at least one of a seat, a headrest, a vehicle window, and a pillar, and
wherein a correlation strength between the feature amount of the first region image and the orientation of the first part and a correlation strength between the feature amount of the second region image and the orientation of the first part are in a complementary relationship by a learning model.
2. The detection device according to claim 1 , wherein the feature amount of the second region image is an area of an exposed portion of the region of the vehicle interior.
3. A learning device comprising:
an image segmenter configured to cut out, from one image of a vehicle interior acquired from a first data provision device, a first region image showing at least a portion of a first part of a body of an occupant, and a second region image from said one image showing at least a portion of a region of the vehicle interior around the first part; and
a learning model generator configured to generate a learning model that learns a correlation between an orientation of the second part acquired from a second data provision device and a feature amount of the first region image and a correlation between the orientation of the second part acquired from the second data provision device and a feature amount of the second region image,
wherein the first part is a face or a head of the occupant, and the feature amount of the first region image is a positional relationship between facial landmarks of the occupant,
wherein the region of the vehicle interior is at least one of a seat, a headrest, a vehicle window, and a pillar, and
wherein a correlation strength between the feature amount of the first region image and an orientation of the first part and a correlation strength between the feature amount of the second region image and the orientation of the first part are in a complementary relationship by a learning model.
4. A detection method performed by a detection device, the detecting method comprising:
cutting out, from one image of a vehicle interior acquired from an imaging device, a first region image showing at least a portion of a first part of a body of an occupant, and a second region image from said one image showing at least a portion of a region of the vehicle interior around the first part; and
detecting an orientation of the first part of the body of the occupant based on a feature amount of the first region image and a feature amount of the second region image,
wherein the first part is a face or a head of the occupant, and the feature amount of the first region image is a positional relationship between facial landmarks of the occupant,
wherein the region of the vehicle interior is at least one of a seat, a headrest, a vehicle window, and a pillar, and
wherein a correlation strength between the feature amount of the first region image and the orientation of the first part and a correlation strength between the feature amount of the second region image and the orientation of the first part are in a complementary relationship by a learning model.
5. A learning method performed by a learning device, the learning method comprising:
cutting out, from one image of a vehicle interior acquired from a first data provision device, a first region image showing at least a portion of a first part of a body of an occupant, and a second region image from said one image showing at least a portion of a region of the vehicle interior around the first part; and
generating a learning model that learns a correlation between an orientation of the second part acquired from a second data provision device and a feature amount of the first region image and a correlation between the orientation of the second part acquired from the second data provision device and a feature amount of the second region image,
wherein the first part is a face or a head of the occupant, and the feature amount of the first region image is a positional relationship between facial landmarks of the occupant,
wherein the region of the vehicle interior is at least one of a seat, a headrest, a vehicle window, and a pillar, and
wherein a correlation strength between the feature amount of the first region image and an orientation of the first part and a correlation strength between the feature amount of the second region image and the orientation of the first part are in a complementary relationship by a learning model.
6. A non-transitory recording medium storing a detection program to be executed by a computer included in a detection device, the detection program causing the computer to execute:
a process of cutting out, from one image of a vehicle interior acquired from an imaging device, a first region image showing at least a portion of a first part of a body of an occupant, and a second region image from said one image showing at least a portion of a region of the vehicle interior around the first part; and
a process of detecting an orientation of the first part of the body of the occupant based on a feature amount of the first region image and a feature amount of the second region image,
wherein the first part is a face or a head of the occupant, and the feature amount of the first region image is a positional relationship between facial landmarks of the occupant,
wherein the region of the vehicle interior is at least one of a seat, a headrest, a vehicle window, and a pillar, and
wherein a correlation strength between the feature amount of the first region image and the orientation of the first part and a correlation strength between the feature amount of the second region image and the orientation of the first part are in a complementary relationship by a learning model.
7. A non-transitory recording medium storing a learning program to be executed by a computer included in a learning device, the learning program causing the computer to execute:
a process of cutting out, from one image of a vehicle interior acquired from a first data provision device, a first region image showing at least a portion of a first part of a body of an occupant, and a second region image from said one image showing at least a portion of a region of the vehicle interior around the first part; and
a process of generating a learning model that learns a correlation between an orientation of the second part acquired from a second data provision device and a feature amount of the first region image and a correlation between the orientation of the second part acquired from the second data provision device and a feature amount of the second region image,
wherein the first part is a face or a head of the occupant, and the feature amount of the first region image is a positional relationship between facial landmarks of the occupant,
wherein the region of the vehicle interior is at least one of a seat, a headrest, a vehicle window, and a pillar, and
wherein a correlation strength between the feature amount of the first region image and an orientation of the first part and a correlation strength between the feature amount of the second region image and the orientation of the first part are in a complementary relationship by a learning model.
8. A detection device comprising:
an image segmenter configured to cut out, from one image of a vehicle interior acquired from an imaging device, a first region image showing at least a portion of a first part of a body of an occupant, and a second region image from said one image showing at least a portion of a second part of the body of the occupant that does not include the first part; and
a detector configured to detect an orientation of the first part of the body of the occupant based on a feature amount of the first region image and a feature amount of the second region image,
wherein the first part is a face or a head of the occupant, and the feature amount of the first region image is a positional relationship between facial landmarks of the occupant,
wherein the second part is one of a shoulder, a chest, and an upper body, and
wherein a correlation strength between the feature amount of the first region image and the orientation of the first part and a correlation strength between the feature amount of the second region image and the orientation of the first part are in a complementary relationship by a learning model.
9. The detection device according to claim 8 , wherein the feature amount of the second region image is an area of the second part in the second region image.
10. A learning device comprising:
an image segmenter configured to cut out, from one image of a vehicle interior acquired from a first data provision device, a first region image showing at least a portion of a first part of a body of an occupant, and a second region image from said one image showing at least a portion of a second part of the body of the occupant that does not include the first part; and
a learning model generator configured to generate a learning model that learns a correlation between an orientation of the second part acquired from a second data provision device and a feature amount of the first region image and a correlation between the orientation of the second part acquired from the second data provision device and a feature amount of the second region image,
wherein the first part is a face or a head of the occupant, and the feature amount of the first region image is a positional relationship between facial landmarks of the occupant,
wherein the second part is one of a shoulder, a chest, and an upper body, and
wherein a correlation strength between the feature amount of the first region image and an orientation of the first part and a correlation strength between the feature amount of the second region image and the orientation of the first part are in a complementary relationship by a learning model.
11. A detection method performed by a detection device, the detecting method comprising:
cutting out, from one image of a vehicle interior acquired from an imaging device, a first region image showing at least a portion of a first part of a body of an occupant, and a second region image from said one image showing at least a portion of a second part of the body of the occupant that does not include the first part; and
detecting an orientation of the first part of the body of the occupant based on a feature amount of the first region image and a feature amount of the second region image,
wherein the first part is a face or a head of the occupant, and the feature amount of the first region image is a positional relationship between facial landmarks of the occupant,
wherein the second part is one of a shoulder, a chest, and an upper body, and
wherein a correlation strength between the feature amount of the first region image and the orientation of the first part and a correlation strength between the feature amount of the second region image and the orientation of the first part are in a complementary relationship by a learning model.
12. A learning method performed by a learning device, the learning method comprising:
cutting out, from one image of a vehicle interior acquired from a first data provision device, a first region image showing at least a portion of a first part of a body of an occupant, and a second region image from said one image showing at least a portion of a second part of the body of the occupant that does not include the first part; and
generating a learning model that learns a correlation between an orientation of the second part acquired from a second data provision device and a feature amount of the first region image and a correlation between the orientation of the second part acquired from the second data provision device and a feature amount of the second region image,
wherein the first part is a face or a head of the occupant, and the feature amount of the first region image is a positional relationship between facial landmarks of the occupant,
wherein the second part is one of a shoulder, a chest, and an upper body, and
wherein a correlation strength between the feature amount of the first region image and an orientation of the first part and a correlation strength between the feature amount of the second region image and the orientation of the first part are in a complementary relationship by a learning model.
13. A non-transitory recording medium storing a detection program to be executed by a computer included in a detection device, the detection program causing the computer to execute:
a process of cutting out, from one image of a vehicle interior acquired from an imaging device, a first region image showing at least a portion of a first part of a body of an occupant, and a second region image from said one image showing at least a portion of a second part of the body of the occupant that does not include the first part; and
a process of detecting an orientation of the first part of the body of the occupant based on a feature amount of the first region image and a feature amount of the second region image,
wherein the first part is a face or a head of the occupant, and the feature amount of the first region image is a positional relationship between facial landmarks of the occupant,
wherein the second part is one of a shoulder, a chest, and an upper body, and
wherein a correlation strength between the feature amount of the first region image and the orientation of the first part and a correlation strength between the feature amount of the second region image and the orientation of the first part are in a complementary relationship by a learning model.
14. A non-transitory recording medium storing a learning program to be executed by a computer included in a learning device, the learning program causing the computer to execute:
a process of cutting out, from one image of a vehicle interior acquired from a first data provision device, a first region image showing at least a portion of a first part of a body of an occupant, and a second region image from said one image showing at least a portion of a second part of the body of the occupant that does not include the first part; and
a process of generating a learning model that learns a correlation between an orientation of the second part acquired from a second data provision device and a feature amount of the first region image and a correlation between the orientation of the second part acquired from the second data provision device and a feature amount of the second region image,
wherein the first part is a face or a head of the occupant, and the feature amount of the first region image is a positional relationship between facial landmarks of the occupant,
wherein the second part is one of a shoulder, a chest, and an upper body, and
wherein a correlation strength between the feature amount of the first region image and an orientation of the first part and a correlation strength between the feature amount of the second region image and the orientation of the first part are in a complementary relationship by a learning model.