Back to results
US 11,900,651 B2Granted
Imaging device and tracking method
AssigneeOlympus Corporation
InventorHisashi Yoneyama
Life of this patent
Estimating the term…
Abstract
An imaging device, comprising an image sensor that exposes a subject image and repeatedly outputs image signals resulting from having subjected this subject image to photoelectric conversion at a fixed period, a subject detection circuit in which the image signals are input to a neural network circuit that has learned operation parameters for detecting a specified subject by deep learning, and that detects the specified subject, and a subject association determination circuit for forming associations based on a positional relationship between a subject (whole subject) that has been detected by the subject detection circuit, and parts of the subject, wherein the specified subject is a subject (whole subject) and parts that have been subjected to subject association.
Claims20
An imaging device, comprising:
- an image sensor that exposes a subject image and repeatedly outputs image signals resulting from having subjected this subject image to photoelectric conversion at a fixed period;
- a subject detection circuit, in which the image signals are input to a neural network circuit that has learned operation parameters for detecting a specified subject by deep learning, that detects the specified subject;
- a subject association determination circuit for forming associations based on a positional relationship between a subject that has been detected by the subject detection circuit, and parts of the subject,
- wherein the specified subject is a subject and parts;
- and
- a positional relationship conditions circuit that determines a position relationship for the subject association determination circuit, wherein
- the positional relationship conditions circuit performs judgment with a central position of a parts detection region that has been detected by the subject detection circuit being within a detection range of the subject that has been detected by the subject detection circuit, or a distance between centers of the subject and the parts detection region being in a specified range, as conditions, and
- the positional relationship conditions circuit widens, or narrows, a range of a subject detection region in accordance with classification of the subject that has been detected by the subject detection circuit.
5 more elements
An imaging device comprising:
- an image sensor that exposes a subject image and repeatedly outputs image signals resulting from having subjected this subject image to photoelectric conversion at a fixed period;
- a subject detection circuit, in which the image signals are input to a neural network circuit that has learned operation parameters for detecting a specified subject by deep learning, that detects the specified subject;
- a subject detection and interpolation circuit that interpolates tracking position, based on feature amounts of the subject, for images that are acquired from the image signals;
- a subject tracking circuit that tracks a subject that has been designated in advance, in images acquired from the image signals;
- and
- a subject tracking position correction circuit that corrects tracking position using results of having subjected a subject to interpolative tracking using the subject detection and interpolation circuit.
3 more elements
A tracking method, comprising:
- exposing a subject image and repeatedly outputting image signals resulting from having subjected this subject image to photoelectric conversion at a fixed period;
- detecting the specified subject using a neural network circuit that has learned operation parameters for detecting the subject, from the image signals;
- forming associations based on a positional relationship between body portions and part portions, in the specified subject that has been detected, wherein
- the specified subject is a subject and parts;
- determining a position relationship;
- performing judgment with a central position of a parts detection region being within a detection range of the subject, or a distance between centers of the subject and the parts detection region being in a specified range, as conditions;
- and
- widening, or narrowing, a range of a subject detection region in accordance with classification of the subject.
5 more elements
A tracking method, comprising:
- exposing a subject image and repeatedly outputting image signals resulting from having subjected this subject image to photoelectric conversion at a fixed period;
- inputting the image signals to a neural network circuit that has learned operation parameters for detecting the specified subject, and detecting the specified subject;
- interpolating tracking position, based on feature amounts of the subject, for images that are acquired from the image signals;
- tracking a subject that has been designated in advance, in images acquired from the image signals;
- and
- correcting tracking position using results of having subjected a subject to interpolative tracking using the subject detection and interpolation circuit.
3 more elements
A non-transitory computer-readable medium storing a computer program for implementing a tracking method, the tracking method comprising:
- exposing a subject image and repeatedly outputting image signals that have been subjected to photoelectric conversion at a fixed period,
- detecting the specified subject using a neural network circuit that has learned operation parameters for detecting the subject, from the image signals;
- forming associations based on a positional relationship between body portions and part portions, in the specified subject that has been detected, wherein
- the specified subject is a subject and parts;
- determining a position relationship;
- performing judgment with a central position of a parts detection region being within a detection range of the subject, or a distance between centers of the subject and the parts detection region being in a specified range, as conditions;
- and
- widening, or narrowing, a range of a subject detection region in accordance with classification of the subject.
5 more elements
A non-transitory computer-readable medium storing a computer program for implementing a tracking method, the tracking method comprising:
- exposing a subject image and repeatedly outputting image signals that have been subjected to photoelectric conversion at a fixed period,
- detecting the specified subject using a neural network circuit that has learned operation parameters for detecting the subject, from the image signals;
- interpolating tracking position, based on feature amounts of the subject, for images that are acquired from the image signals;
- tracking a subject that has been designated in advance, in images acquired from the image signals;
- and
- correcting tracking position using results of having subjected a subject to interpolative tracking using the subject detection and interpolation circuit.
3 more elements
Dependencies are inferred from the claim text.