Abstract: Methods, systems, and apparatus for motion-based human video detection are disclosed. A method includes generating a representation of a difference between two frames of a video; providing, to an object detector, a particular frame of the two frames and the representation of the difference between two frames of the video; receiving an indication that the object detector detected an object in the particular frame; determining that detection of the object in the particular frame was a false positive detection; determining an amount of motion energy where the object was detected in the particular frame; and training the object detector based on penalization of the false positive detection in accordance with the amount of motion energy where the object was detected in the particular frame.

Abstract: A detection device includes an acquirer that acquires a video of each of a plurality of bearings of a structure including the plurality of bearings, an extractor that extracts a dynamic feature corresponding to a plurality of degrees of freedom of each of the plurality of bearings based on the video, and an identifier that identifies, among the plurality of bearings, a bearing whose dynamic feature fails to match a dynamic feature of one or more other bearings of the plurality of bearings.

Abstract: In one embodiment, a first deep fusion reasoning engine (DFRE) agent in a network receives first sensor data from a first set of one or more sensors in the network. The first DFRE agent translates the first sensor data into symbolic data. The first DFRE agent applies, using a symbolic knowledge base maintained by the first DFRE agent, symbolic reasoning to the symbolic data to make an inference regarding the first sensor data. The first DFRE agent updates, based on the inference regarding the first sensor data, the knowledge base. The first DFRE agent propagates the inference to one or more other DFRE agents in the network.

Abstract: A method for virtually representing human body poses includes receiving positioning data detailing parameters of one or more body parts of a human user based at least in part on input from one or more sensors. One or more mapping constraints are maintained that relate a model articulated representation to a target articulated representation. A model pose of the model articulated representation and a target pose of the target articulated representation are concurrently estimated based at least in part on the positioning data and the one or more mapping constraints. The previously-trained pose optimization machine is trained with training positioning data having ground truth labels for the model articulated representation. The target articulated representation is output for display with the target pose as a virtual representation of the human user.

Abstract: Methods, apparatus and systems for wireless motion recognition are described. In one example, a described system comprises: a transmitter configured for transmitting a first wireless signal through a wireless multipath channel of a venue; a receiver configured for receiving a second wireless signal through the wireless multipath channel; and a processor. The second wireless signal differs from the first wireless signal due to the wireless multipath channel that is impacted by a motion of an object in the venue. The processor is configured for: obtaining a time series of channel information (TSCI) of the wireless multipath channel based on the second wireless signal, tracking the motion of the object based on the TSCI to generate a gesture trajectory of the object, and determining a gesture shape based on the gesture trajectory and a plurality of pre-determined gesture shapes.

Abstract: The invention relates to a process for monitoring vehicles on a road by a system comprising at least one radar sensor and a second sensor different from the radar sensor, wherein the second remote sensor is a time-of-flight optical sensor or optical image sensor, the process comprising a temporal readjustment and a spatial matching in order to obtain a set of measurement points each assigned to first characteristics derived from the radar data and second characteristics derived from the optical data, the determination of the radar vehicle trackings and of the optical vehicle trackings, a comparison of similarity between the radar vehicle trackings and the optical vehicle trackings, the elimination of the radar vehicle trackings for which no optical vehicle tracking is similar, the process comprising monitoring a parameter derived from first characteristics of a retained radar vehicle tracking.

Abstract: A video processing method includes dividing a region of a current frame to obtain a plurality of image blocks, obtaining a historical motion information candidate list, and obtaining candidate historical motion information for the plurality of image blocks according to the historical motion information candidate list. The candidate historical motion information is a candidate in the historical motion information candidate list. The method further includes performing prediction for the plurality of image blocks according to the candidate historical motion information. A size of each of the plurality of image blocks is smaller than or equal to a preset size. The same historical motion information candidate list is used for the plurality of image blocks during the prediction. The historical motion information candidate list is not updated while the prediction is being performed for the plurality of image blocks.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for selecting locations in an environment of a vehicle where objects are likely centered and determining properties of those objects. One of the methods includes receiving an input characterizing an environment external to a vehicle. For each of a plurality of locations in the environment, a respective first object score that represents a likelihood that a center of an object is located at the location is determined. Based on the first object scores, one or more locations from the plurality of locations are selected as locations in the environment at which respective objects are likely centered. Object properties of the objects that are likely centered at the selected locations are also determined.

Abstract: A method includes receiving a series of road images from a side-view camera sensor of the autonomous driving vehicle. For each object from objects captured in the series of road images, a series of bounding boxes in the series of road images is generated, and a direction of travel or stationarity of the object is determined. The methods and apparatus also include determining a speed of each object for which the direction of travel has been determined and determining, based on the directions of travel, speeds, or stationarity of the objects, whether the autonomous driving vehicle can safely move in a predetermined direction. Furthermore, one or more control signals is sent to the autonomous driving vehicle to cause the autonomous driving vehicle to move or to remain stationary based on determining whether the autonomous driving vehicle can safely move in the predetermined direction.

Abstract: A plurality of motion vectors are acquired from consecutive images. From the acquired plurality of motion vectors, a motion vector of interest and neighboring motion vectors neighboring the motion vector of interest are selected and a degree of similarity between two motion vectors is acquired. A value related to the total number of neighboring motion vectors having high degrees of similarity to the motion vector of interest is acquired as a degree of reliability.

Abstract: The present disclosure relates to a narrow slit channel visualization experimental device and method under a six-degree-of-freedom motion condition. The system comprises a six-degree-of-freedom motion simulation platform, a main circulation loop, a cooling water system, an electric heating system and a bubble monitoring system, wherein the main circulation loop is composed of an S-shaped preheater, a three-surface visualization experimental section, a double-pipe condenser, a pressurizing circulating pump, a voltage stabilizer and related equipment, wherein the cooling water system is composed of the double-pipe condenser, a plate heat exchanger, a cooling tower, a cooling fan, a cooling water tank and related equipment, wherein the electric heating system is composed of a direct-current power supply, a low-voltage power controller and a transformer, and wherein the bubble monitoring system is composed of two high-speed cameras, a PIV measuring system and an electric servo module.

Abstract: Disclosed is a computer-implemented of adapting a biomechanical model of an anatomical body part of a patient to a current status of the patient. The method encompasses determination of a currently executed step of a workflow such as a medical intervention, the result of the determination serving as a basis for adapting and/or updating a biomechanical model of an anatomical body part to the corresponding current status of the patient. The determination of the current workflow step may also be used as basis for controlling an imaging device for tracking entities around the patient or for imaging the anatomical body part or acquiring further data or for urging the user to perform a specific action such as acquisition of information using a tracked instrument such as a pointer. The biomechanical model has been generated from atlas data.

Abstract: An imaging apparatus acquires pixel signals for each predetermined-size partial region of an imaging plane and for detecting a region among the regions of the imaging plane in which a subject image has changed, the number of acquired pixel signals being less than the number of pixels included in the partial region, and controls to perform image capturing in response to detection of the region in which the subject image has changed, and to output a captured image based the pixel signals. The imaging apparatus acquires pixel signals from each partial region of the imaging plane by, when pixel signals are to be acquired from a first partial region of the imaging plane, acquiring pixel signals from a second partial region that is adjacent to the first partial region and partially overlapped with the first partial region.

Abstract: Various methods are provided for the generation of motion vectors in the context of 3D computer-generated images. In one example, a method includes generating, for each pixel of one or more objects to be rendered in a current frame, a 1-phase motion vector (MV1) and a 0-phase motion vector (MV0), each MV1 and MV0 having an associated depth value, to thereby form an MV1 texture and an MV0 texture, each MV0 determined based on a camera MV0 and an object MV0, converting the MV1 texture to a set of MV1 pixel blocks and converting the MV0 texture to a set of MV0 pixel blocks and outputting the set of MV1 pixel blocks and the set of MV0 pixel blocks for image processing.

Abstract: Methods, systems and apparatuses may provide for technology that selects a player from a plurality of players based on an automated analysis of two-dimensional (2D) video data associated with a plurality of cameras, wherein the selected player is nearest to a projectile depicted in the 2D video data. The technology may also track a location of the selected player over a subsequent plurality of frames in the 2D video data and estimate a location of the projectile based on the location of the selected player over the subsequent plurality of frames.

Abstract: A computer-implemented method of object enhancement in endoscopy images includes capturing an image of an object within a surgical operative site via an imaging device, determining a size of the object based on the captured image of the object, displaying the captured image of the object, and displaying on the displayed captured image of the object a representation of the determined size of the object.

Abstract: Device, methods and systems for implementing aspects of orthogonal time frequency space (OTFS) modulation in wireless systems are described. In an aspect, the device may include a surface of an object for receiving an electromagnetic signal. The surface may be structured to perform a non-electrical function for the object. The surface may generate an electrical signal from an electromagnetic signal. The electromagnetic signal may be received from a transmitter. The transmitter may map digital data to a digital amplitude modulation constellation in a time-frequency space. The digital amplitude modulation constellation may be mapped to a delay-Doppler domain and the transmitter may transmit to the surface according to an orthogonal time frequency space modulation signal scheme. The apparatus may further include a demodulator to demodulate the electrical signal to determine digital data.

Abstract: A skeleton model updating apparatus, a skeleton model updating method, and a program by which time and effort for changing the pose of a skeleton model to a known standard pose can be reduced. A target node identifying section identifies target nodes from among a plurality of nodes included in a skeleton model that is in a pose other than a standard pose. A reference node identifying section identifies a reference node that is positioned closest to the side of the target nodes, from among nodes that are connected to all of the target nodes via one or more bones. A position deciding section decides positions of the target nodes such that relative positions of the target nodes with respect to the position of the reference node are adjusted to predetermined positions.

Abstract: A distance measurement camera contains a first optical system OS1 for forming a first subject image, a second optical system OS2 for forming a second subject image, an imaging part S for imaging the first subject image and the second subject image and a distance calculating part 4 for performing calculation depending on image heights of distance measurement target points of the first subject image and the second subject image corresponding to a distance measurement target point of a subject 100 to calculate the distance to the subject 100 based on an image magnification ratio between a magnification of the first subject image and a magnification of the first subject image and a magnification of the second subject image.

Abstract: Systems, methods, tangible non-transitory computer-readable media, and devices for detecting objects are provided. For example, the disclosed technology can obtain a representation of sensor data associated with an environment surrounding a vehicle. Further, the sensor data can include sensor data points. A point classification and point property estimation can be determined for each of the sensor data points and a portion of the sensor data points can be clustered into an object instance based on the point classification and point property estimation for each of the sensor data points. A collection of point classifications and point property estimations can be determined for the portion of the sensor data points clustered into the object instance. Furthermore, object instance property estimations for the object instance can be determined based on the collection of point classifications and point property estimations for the portion of the sensor data points clustered into the object instance.

Abstract: Systems and methods for predicting object motion and controlling autonomous vehicles are provided. In one example embodiment, a computer implemented method includes obtaining state data indicative of at least a current or a past state of an object that is within a surrounding environment of an autonomous vehicle. The method includes obtaining data associated with a geographic area in which the object is located. The method includes generating a combined data set associated with the object based at least in part on a fusion of the state data and the data associated with the geographic area in which the object is located. The method includes obtaining data indicative of a machine-learned model. The method includes inputting the combined data set into the machine-learned model. The method includes receiving an output from the machine-learned model. The output can be indicative of a plurality of predicted trajectories of the object.

Abstract: Systems and methods for generating a panorama image. Captured images are coarsely aligned, and then finely aligned based on a combination of constraint values. The panorama image is generated from the finely aligned images.

Abstract: A medication tray may include a boundary wall defining a boundary outline. Markers may be carried by the boundary wall. A mobile device may obtain a first image of the medication tray, and determine a number of the markers from the first image. When the number equals or exceeds a threshold, an edge detection algorithm may be applied to the first image to identify the boundary outline based upon the markers, and generate a stocking list of the tray based upon the outline. When the determined number does not exceed the threshold, the device may obtain a second image of the tray, and determine an updated number of the markers from the first and second images. When the updated number equals or exceeds the threshold, the algorithm may be applied to the images to identify the outline based upon the markers, and generate the stocking list based upon the outline.

Abstract: An imaging apparatus acquires pixel signals for each predetermined-size partial region of an imaging plane and for detecting a region among the regions of the imaging plane in which a subject image has changed, the number of acquired pixel signals being less than the number of pixels included in the partial region, and controls to perform image capturing in response to detection of the region in which the subject image has changed, and to output a captured image based the pixel signals. The imaging apparatus acquires pixel signals from each partial region of the imaging plane by, when pixel signals are to be acquired from a first partial region of the imaging plane, acquiring pixel signals from a second partial region that is adjacent to the first partial region and partially overlapped with the first partial region.

Abstract: A system (11) includes a medical probe (36) for insertion into a cavity of an organ, which includes a position and direction sensor (60) and a camera (45), both operating in a sensor coordinate system (62). The system further includes a processor (44) configured to: receive, from an imaging system (21) operating in an image coordinate system (28), a three-dimensional image of the cavity including open space and tissue; receive, from the medical probe, signals indicating positions and respective directions of the medical probe inside the cavity; receive, from the camera, respective visualized locations inside the cavity; register the image coordinate system with the sensor coordinate system so as to identify one or more voxels in the image at the visualized locations, and when the identified voxels have density values in the received image that do not correspond to the open space, to update the density values of the identified voxels to correspond to the open space.

Abstract: An element recognition apparatus (100) segments three-dimensional skeleton information including a subject performing a series of performances, acquired in chronological order, into a plurality of units in accordance with a predetermined rule. The element recognition apparatus (100) determines whether a postural motion corresponding to a first unit among the units is a first motion having a feature indicating that the subject is stationary or a second motion different from the first motion. The element recognition apparatus (100) determines, based on a determination result corresponding to the first unit and a recognition result of recognition on the type of moving motion in a second unit that is continuous with the first unit, an element corresponding to the combination including at least the first unit and the second unit.

Abstract: A video indexing system identifies groups of frames within a video frame sequence captured by a static camera during a same scene. Context metadata is generated for each frame in each group based on an analysis of fewer than all frames in the group. The frames are indexed in a database in association with the generated context metadata.

Abstract: A method for tracking movement of nematode worms comprises: (i) providing a plurality of worms on a translucent substrate; (ii) obtaining a first image of a field of view including the plurality of worms by transmission imaging; (iii) obtaining a first difference image of the plurality of worms corresponding to an intensity difference between said first image and a background image of the field of view; (iv) repeating the following steps (a) to (d) a plurality of N times, for n=1 to N: (a) determining, from the first difference image, an nth pixel corresponding to a maximum intensity difference; (b) selecting, from the first difference image, an nth block of pixels comprising the selected nth pixel; (c) determining a coordinate associated with the selected nth block of pixels; and (d) updating said first difference image by setting each pixel of said nth block of pixels in said first difference image to a value corresponding to a zero or low intensity difference; (v) obtaining a sequence of M subsequent image

Abstract: Conventionally, activity detection has been through one mode i.e., smart watch. Though it works in reasonable cases, there are chances of false positives considerably. Other approaches include surveillance which limits itself to object detection. Embodiments of present disclosure provide systems and methods for detecting activities performed by user from data captured from multiple sensors. A first input (FI) comprising accelerometer data, heart rate and gyroscope data and second input (SI) comprising video data are obtained. Features are extracted from FI and pre-processed for a first activity (FA) detection using activity prediction model. Frames from SI are processed for creating bounding box of user and resized thereof to extract pose coordinates vector. Distance between vector of pose coordinates and training vectors of pose coordinates stored in the system is computed and a second activity (SA) is detected accordingly. Both the FA and SA are validated for determining true and/or false positive.

Abstract: A computer-implemented method, a computer system and a computer program product improve the quality of multimedia. The method includes displaying a current frame of a video. The method also includes generating dataframes for the current frame and for a reference frame of the video. The method further includes comparing the dataframes for the reference and current frames. In addition, the method includes determining a quality metric of the current frame based on the comparison of the dataframes for the reference and current frames. Finally, the method includes altering an orientation of the display of the current frame in response to determining that the quality metric of the current frame is below a threshold.

Abstract: Embodiments and aspects described herein provide methods and systems for determining pressure difference across a tube arising from fluid flow within the tube, comprising: obtaining three-dimensional time dependent fluid velocity data at a plurality of points along the tube; processing the three-dimensional time dependent fluid velocity data to determine: i) a flow rate (Q) of the fluid through the tube; ii) the kinetic energy (K) of the fluid flow through the tube; iii) an advective energy rate (A) of the fluid flow through the tube; and iv) a viscous dissipation rate (V) pertaining to the fluid flow; and calculating the pressure difference in dependence on all of the flow rate (Q), kinetic energy (K), advective energy rate (A), and viscous dissipation rate (V). Further embodiments are also described.

Abstract: As learning data, an image of a virtual space corresponding to a physical space and geometric information of the virtual space is generated. Learning processing of a learning model is performed using the learning data. A position and/or orientation of an image capturing device is calculated based on geometric information output from the learning model when a captured image of the physical space captured by the image capturing device is input to the learning model.

Abstract: An image processing apparatus detects a tracking target object in an image, executes tracking processing to track the object, determines whether an attribute of an object detected from the image is a predetermined attribute, identifies, when a first state in which the object is detected changes to a second state with the object not detected, a given object included in the image and positioned at least partially in front of the object in the second state, based on a position of the object in the first state, controls the tracking processing, based on a determination whether an attribute of the given object is the predetermined attribute, and determines, based on the result, whether to continue the tracking processing. The tracking processing continues until at least a predetermined time has elapsed when a determination to continue the tracking processing on the tracking target object for the predetermined time has been made.

Abstract: A vehicle dash cam may be configured to execute one or more neural networks (and/or other artificial intelligence), such as based on input from one or more of the cameras and/or other sensors associated with the dash cam, to intelligently detect safety events in real-time. Detection of a safety event may trigger an in-cab alert to make the driver aware of the safety risk. The dash cam may include logic for determining which asset data to transmit to a backend server in response to detection of a safety event, as well as which asset data to transmit to the backend server in response to analysis of sensor data that did not trigger a safety event. The asset data transmitted to the backend server may be further analyzed to determine if further alerts should be provided to the driver and/or to a safety manager.

Abstract: Systems and methods for generating motion forecast data for actors with respect to an autonomous vehicle and training a machine learned model for the same are disclosed. The computing system can include an object detection model and a graph neural network including a plurality of nodes and a plurality of edges. The computing system can be configured to input sensor data into the object detection model; receive object detection data describing the location of the plurality of the actors relative to the autonomous vehicle as an output of the object detection model; input the object detection data into the graph neural network; iteratively update a plurality of node states respectively associated with the plurality of nodes; and receive, as an output of the graph neural network, the motion forecast data with respect to the plurality of actors.

Abstract: Provided are embodiments including a system for automatically generating a plan of scan locations for performing a scanning operation where the system includes a storage medium that is coupled to a processor. The processor is configured to receive a map of an environment, apply a distance transform to the map, wherein the distance transform determines a path through the map, wherein the path comprises a plurality of points, and identify a set of candidate scan locations based on the path. The processor is also configured to select scan locations from the set of candidate scan locations for performing 3D scans, and perform the 3D scans of the environment based on the selected scan locations. Also provided are embodiments for a method and computer program product for automatically generating a plan of scan locations for performing a scanning operation.

Abstract: In the case where a three-dimensional position of a moving body is calculated based on image data taken by a plurality of cameras that are synchronized, high performance equipment and system such as a system to make the synchronization among the plurality of cameras and a camera having a built-in function to make the synchronization are required. It is also required to fix the camera position with high accuracy beforehand. It is made possible to calculate a trajectory of a moving body as a target in the three-dimensional space using image data taken by a plurality of cameras that are non-synchronized mutually, thereby solving the above issue. And positions of respective cameras are calculated in the three-dimensional space from a plurality of reference points having fixed position coordinates in the three-dimensional space that are commonly shown in the image data of the respective cameras, thereby solving the above issue.

Abstract: This disclosure relates to systems and methods of obtaining accurate motion and orientation estimates for a vehicle traveling at high speed based on images of a road surface. A purpose of these systems and methods is to provide a supplementary or alternative means of locating a vehicle on a map, particularly in cases where other locationing approaches (e.g., GPS) are unreliable or unavailable.

Abstract: A system configured to assist a user in scanning a physical environment in order to generate a three-dimensional scan or model. In some cases, the system may include an interface to assist the user in capturing data usable to determine a scale or depth of the physical environment and to perform a scan in a manner that minimizes gaps.

Abstract: A camera height calculation method that causes a computer to execute a process, the process includes obtaining one or more images captured by an in-vehicle camera, extracting one or more feature points from the one or more images, identifying first feature points that exist over a road surface from the one or more feature points, and calculating a height of the in-vehicle camera from the road surface, based on positions of the identified first feature points.

Abstract: The present disclosure relates to a motion estimation method, a chip, an electronic device, and a storage medium. The present disclosure is beneficial to improving the accuracy of motion estimation.

Abstract: A three-dimensional accelerometer in an animal tag registers a first set of acceleration parameters expressing a respective acceleration of the tag along each of three independent spatial axes. A processor in the tag derives a respective estimated gravity-related component in each parameter in the first set, and compensates for the respective estimated gravity-related components in the first set to obtain a second set of acceleration parameters representing respective accelerations of the animal tag along each of three independent spatial axes each in which the parameter is balanced around a base level with no influence of gravitation. The processor determines behavior-related data of rise-up and/or lie-down movements of an animal carrying the animal tag based on deviations in a single parameter in the second set of acceleration parameters relative to the base level.

Abstract: Example implementations described herein involve systems and methods for a mobile application device to playback and record augmented reality (AR) overlays indicating gestures to be made to a recorded device screen. A device screen is recorded by a camera of the mobile device, wherein a mask is overlaid on a user hand interacting with the device screen. Interactions made to the device screen are detected based on the mask, and AR overlays are generated corresponding to the reactions.

Abstract: Poses of a person depicted within video frame may be determined. The poses of the person may be used to generate intermediate video frames between the video frames.

Abstract: The shape or movement of a gesturing body or portion thereof in two- or three-dimensional space is ascertained from the path of the outline of the shape or the path of the movement. The method disclosed involves receiving input of data that represents a path and using artificial intelligence to recognize the meaning of the path, i.e., to recognize which of a plurality of pre-prepared meanings is the meaning of a gesture. As pre-processing for inputting location data for a point group along the path to the artificial intelligence, at least one attribute from among the location, size, and direction of the entire point group is extracted, and location data for the point group is converted to attribute invariant location data that is relative to the extracted attribute(s) but not dependent on the extracted attribute(s). Then data that includes the attribute invariant location data and the extracted attribute(s) is inputted to the artificial intelligence as input data.

Abstract: Methods and systems are provided for detecting patient motion during a diagnostic scan. In one example, a method for a medical imaging system includes obtaining output from one or more patient monitoring devices configured to monitor a patient before and during a diagnostic scan executed with the medical imaging system, receiving a request to initiate the diagnostic scan, tracking patient motion based on the output from the one or more patient monitoring devices, and initiating the diagnostic scan responsive to patient motion being below a threshold level.

Abstract: A method is for providing a medical image of a patient, acquired via a computed tomography apparatus. An embodiment of the method includes acquiring first projection data of a first measurement region; acquiring second projection data of a second measurement region; registering a reference image to the at least one respiration-correlated image of the patient, wherein the reference image corresponds to the at least one functional image of the patient or is reconstructed under a second reconstruction rule from the second projection data, to produce a deformation model; applying the deformation model to the at least one functional image of the patient; combining the at least one functional image of the patient, deformed by the applying of the deformation model, with the at least one respiration-correlated image of the patient, to produce the medical image of the patient; and providing the medical image of the patient.

Abstract: A computing device captures a live video of a user. For a first frame of the live video, the computing device obtains first target positional coordinates of a first target point located a first predetermined distance from the computing device and obtains first background data. For a second frame, the computing device obtains second target positional coordinates of a second target point located a second predetermined distance from the computing device and obtains second background data. The computing device calculates a target motion vector based on the first target point and the second target point and calculates a background motion vector based on feature points in the first background data and the second background data. The computing device determines a difference value between the target motion vector and the background motion vector and determines whether the user is spoofing the computing device based on the difference value.

Abstract: A moving state analysis device improves accuracy of moving state recognition by including a detection unit configured to detect, from image data associated with a frame, an object and a region of the object, for each of frames that constitute first video data captured in a course of movement of a first moving body, and a learning unit configured to learn a DNN model that takes video data and sensor data as input and that outputs a probability of each moving state, based on the first video data, a feature of first sensor data measured in relation to the first moving body and corresponding to a capture of the first video data, a detection result of the object and the region of the object, and information that indicates a moving state associated with the first video data.

Abstract: [Problem] To provide an action-estimating device with which an action of a subject appearing in a plurality of time-series images can be precisely estimated. [Solution] In the action-estimating device 1, an estimating-side detecting unit 13 detects a plurality of articulations A appearing in each time-series image Y on the basis of a reference having been stored in an estimating-side identifier 11 and serving to identify the plurality of articulations A. An estimating-side measuring unit 14 measures the coordinates and the depths of the plurality of articulations A appearing in each of the time-series images Y. On the basis of displacement in the plurality of time-series images Y of the measured coordinate and depth of each of the articulations A, a specifying unit 15 specifies, from among the plurality of articulations A, an articulation group B which belongs to a given subject.