摘要:
This posture state estimation device is capable of estimating with high accuracy the posture state of an object. The posture state estimation device (100) is a device for estimating the posture state of an object having a plurality of sections connected by joints on the basis of image data that images the object, and has a section candidate extraction unit (140) for extracting a section candidate for a section from the image data, a complementary section candidate extraction unit (160) which estimates that a portion of an unextracted section for which a section candidate has not been extracted is being shadowed by an already extracted section for which a section candidate has been extracted in order to extract a section candidate of the unextracted section, and a posture state estimation unit (170) for estimating the posture state of the object on the basis of the extracted section candidate.
摘要:
Provided is a person detection device with which it is possible to estimate a state of a part of a person from an image. A person detection device (100) comprises: an evaluation unit (430) which acquires a prescribed outline of a person from an evaluation image; and a shoulder position calculation unit (440) and an orientation estimation unit (500) which estimate a state of a prescribed part of a person which is included in the evaluation image from the prescribed outline of the person which is acquired from the evaluation image, on the basis of an estimation model which denotes a relation between the prescribed outline and the state of the prescribed part of the person.
摘要:
Provided is a posture estimation device with which it is possible to estimate the posture of an object having joints with a high degree of precision. A posture estimation device (100) carries out an estimation of a posture of an object on the basis of image data wherein an object having a plurality of sites which are connected by joints is photographed. The device comprises: a reference model storage unit (140) which stores a reference model, which defines the locations of the sites, for each posture; a weighting computation unit (150) which carries out weighting on the sites for each posture such that the weighting of concentrated parts is reduced; and a posture estimation unit (160) which estimates the posture of the object by applying the weighting and comparing the reference model with the object.
摘要:
The present invention is a posture estimation device for estimating a wide variety of 3-dimensional postures by using a skeletal model. The posture estimation device (200) has: a skeletal backbone estimation unit (230) for estimating the position of a feature location of a person within an acquired image; a location extraction unit (240) which generates a likelihood map indicating the certainty that a location other than the feature location of the person exists in the acquired image based on the position of the feature location of the person; and a skeletal model evaluation unit (250) for evaluating, based on the likelihood map, a candidate group which includes a plurality of 2-dimensional skeletal models as candidates and such that each 2-dimensional skeletal model is configured from a line group representing each location and a point group representing coupling between each location and corresponds to one posture of the person.
摘要:
Disclosed is an object detection method capable of detecting with high precision information relating to a jointed object from image data. An object detection device (160) detects information relating to an object from image data of images captured of an object having multiple parts connected by joints. The disclosed object detection device (160) is provided with a joint angle extraction unit (161) which extracts the angle of a joint connecting two parts from candidates of the positions of two neighboring parts obtained from the image data, and a part length ratio estimation unit (165) which uses the joint angle to perform the detection described above.
摘要:
Provided is a person detection device with which it is possible to estimate a state of a part of a person from an image. A person detection device (100) comprises: an evaluation unit (430) which acquires a prescribed outline of a person from an evaluation image; and a shoulder position calculation unit (440) and an orientation estimation unit (500) which estimate a state of a prescribed part of a person which is included in the evaluation image from the prescribed outline of the person which is acquired from the evaluation image, on the basis of an estimation model which denotes a relation between the prescribed outline and the state of the prescribed part of the person.
摘要:
This posture state estimation device is capable of estimating with high accuracy the posture state of an object. The posture state estimation device (100) is a device for estimating the posture state of an object having a plurality of sections connected by joints on the basis of image data that images the object, and has a section candidate extraction unit (140) for extracting a section candidate for a section from the image data, a complementary section candidate extraction unit (160) which estimates that a portion of an unextracted section for which a section candidate has not been extracted is being shadowed by an already extracted section for which a section candidate has been extracted in order to extract a section candidate of the unextracted section, and a posture state estimation unit (170) for estimating the posture state of the object on the basis of the extracted section candidate.
摘要:
Provided is a posture estimation device with which it is possible to estimate the posture of an object having joints with a high degree of precision. A posture estimation device (100) carries out an estimation of a posture of an object on the basis of image data. An object having a plurality of sites which are connected by joints is photographed. The posture estimation device includes a reference model storage unit (140) which stores a reference model, which defines the locations of the sites, for each posture. A weighting computation unit (150) carries out weighting on the sites for each posture such that the weighting of concentrated parts is reduced. A posture estimation unit (160) estimates the posture of the object by applying the weighting and comparing the reference model with the object.
摘要:
Disclosed is an orientation state estimation device capable of estimating with high accuracy the orientation state of a jointed body. An orientation state estimation device (100) estimates the orientation state of a body on the basis of image data of the body having multiple parts connected by joints. The device is provided with: a likelihood map generation unit (150) which, from the image data, for at least two parts of the jointed body, generates a likelihood map showing the plausibility distribution of where each part is most plausibly positioned; and an orientation state estimation unit (160) which, when a learning likelihood map, which is associated in advance with an orientation state, and an estimated likelihood map, which is generated on the basis of the image data, coincide to a high degree, estimates that the orientation state associated with said learning likelihood map is the orientation state of the object.
摘要:
Disclosed is an orientation state estimation device capable of estimating with high accuracy the orientation state of a jointed body. An orientation state estimation device (100) estimates the orientation state of a body on the basis of image data of the body having multiple parts connected by joints. The device is provided with: a likelihood map generation unit (150) which, from the image data, for at least two parts of the jointed body, generates a likelihood map showing the plausibility distribution of where each part is most plausibly positioned; and an orientation state estimation unit (160) which, when a learning likelihood map, which is associated in advance with an orientation state, and an estimated likelihood map, which is generated on the basis of the image data, coincide to a high degree, estimates that the orientation state associated with said learning likelihood map is the orientation state of the object.