Abstract: A learning data processing unit accepts, as input, a plurality of pieces of learning data for a respective plurality of tasks, and calculates, for each of the tasks, a batch size which meets a condition that a value obtained by dividing a data size of corresponding one of the pieces of learning data by the corresponding batch size is the same between the tasks. A batch sampling unit samples, for each of the tasks, samples from corresponding one of the pieces of learning data with the corresponding batch size calculated by the learning data processing unit. A learning unit updates a weight of a discriminator for each of the tasks, using the samples sampled by the batch sampling unit.

Abstract: A region specifying unit (28) takes, as test data, a plurality of pieces of image data obtained by photographing a photographing region with a photographing device, and specifies an enlarging region in accordance with an appearance number indicating how many objects smaller than a standard size appear in the test data. A data extraction unit (23) extracts, out of the image data obtained by photographing the photographing region, image data of the enlarging region, as partial data. A size modification unit (24) size-modifies the partial data to a request size requested by an object detection model being a model that detects an object from image data. An object detection unit (25) inputs the size-modified partial data to the object detection model, and detects a target object from the partial data.

Abstract: An image analysis device includes a person detecting unit for detecting person images from image information from a surveillance camera, and acquiring both pieces of person image information as a result of the detection of person images, and pieces of attribution information about persons corresponding to the pieces of person image information, and a person narrowing down unit for performing narrowing-down of person image information with respect to the pieces of person image information and the pieces of person attribution information which are acquired by the person detecting unit, on the basis of a person narrowing down rule, to acquire person image information after the narrowing-down and attribution information about a person corresponding to the person image information.

Abstract: An extraction unit (21) treats two or more consecutive frames of a plurality of frames constituting video data as a time-window frame, and extracts a plurality of time-window frames from the video data so as to cause at least one frame to be overlappingly included in adjacent time-window frames. A tracking unit (22) treats each time-window frame as a target time-window frame, and tracks target objects in the target time-window frame to generate a tracking result. An integration unit (23) integrates tracking results generated in individual time-window frames, based on a frame overlappingly included in more than one time-window frame, to generate an integration result.

Abstract: A graph generation unit (21) generates a tracking graph by setting, as a node, each of a plurality of target objects detected in a plurality of frames constituting video data, and creating an edge between two consecutive frames to connect target objects. A vector calculation unit (22) calculates a velocity vector of a target object detected in a target frame, based on a correspondence between the target object detected in the target frame and a target object detected in a previous frame of the target frame. A cost calculation unit (23) calculates a cost of an edge connecting a target object in the target frame and a target object in a next frame of the target frame in the tracking graph. A correspondence determination unit (24) determines a correspondence between a target object in the target frame and a target object in the next frame, based on the tracking graph and the cost.

Abstract: An image analysis device includes a person detecting unit for detecting person images from image information from a surveillance camera, and acquiring both pieces of person image information as a result of the detection of person images, and pieces of attribution information about persons corresponding to the pieces of person image information, and a person narrowing down unit for performing narrowing-down of person image information with respect to the pieces of person image information and the pieces of person attribution information which are acquired by the person detecting unit, on the basis of a person narrowing down rule, to acquire person image information after the narrowing-down and attribution information about a person corresponding to the person image information.

Abstract: An extraction unit (21) treats two or more consecutive frames of a plurality of frames constituting video data as a time-window frame, and extracts a plurality of time-window frames from the video data so as to cause at least one frame to be overlappingly included in adjacent time-window frames. A tracking unit (22) treats each time-window frame as a target time-window frame, and tracks target objects in the target time-window frame to generate a tracking result. An integration unit (23) integrates tracking results generated in individual time-window frames, based on a frame overlappingly included in more than one time-window frame, to generate an integration result.

Abstract: A graph generation unit (21) generates a tracking graph by setting, as a node, each of a plurality of target objects detected in a plurality of frames constituting video data, and creating an edge between two consecutive frames to connect target objects. A vector calculation unit (22) calculates a velocity vector of a target object detected in a target frame, based on a correspondence between the target object detected in the target frame and a target object detected in a previous frame of the target frame. A cost calculation unit (23) calculates a cost of an edge connecting a target object in the target frame and a target object in a next frame of the target frame in the tracking graph. A correspondence determination unit (24) determines a correspondence between a target object in the target frame and a target object in the next frame, based on the tracking graph and the cost.

Abstract: A learning data processing unit accepts, as input, a plurality of pieces of learning data for a respective plurality of tasks, and calculates, for each of the tasks, a batch size which meets a condition that a value obtained by dividing a data size of corresponding one of the pieces of learning data by the corresponding batch size is the same between the tasks. A batch sampling unit samples, for each of the tasks, samples from corresponding one of the pieces of learning data with the corresponding batch size calculated by the learning data processing unit. A learning unit updates a weight of a discriminator for each of the tasks, using the samples sampled by the batch sampling unit.

Abstract: A learning sample collector is configured to collect learning samples which have been classified into respective classes through N-classes discrimination (N is a natural number of 3 or more). A classifier is configured to reclassify the learning samples collected by the learning sample collector into classes applied to M-classes discrimination, where M is smaller than N (M is a natural number of 2 or more and is less than N). A learner is configured to learn a discriminator for performing the M-classes discrimination on a basis of the learning samples reclassified by the classifier.

Abstract: When the signal format of a color image is YUV4:2:2, and an intra prediction parameter, which is used for an intra-frame prediction process on the chrominance signals in a prediction block, shows the same prediction mode as an intra prediction mode for the luminance signal in the prediction block, an intra prediction unit 4 converts an index indicating the intra prediction mode for the luminance signal, and performs the intra-frame prediction process on the chrominance signals in an intra prediction mode indicated by the index after conversion.

Abstract: When the signal format of a color image is YUV4:2:2, and an intra prediction parameter, which is used for an intra-frame prediction process on the chrominance signals in a prediction block, shows the same prediction mode as an intra prediction mode for the luminance signal in the prediction block, an intra prediction unit 4 converts an index indicating the intra prediction mode for the luminance signal, and performs the intra-frame prediction process on the chrominance signals in an intra prediction mode indicated by the index after conversion.