Abstract: Systems and methods are disclosed for lookup-based convolutional neural networks. For example, methods may include applying a convolutional neural network to image data based on an image to obtain an output, in which a layer of the convolutional network includes filters with weights that are stored as a dictionary (D) of channel weight vectors, a respective lookup index tensor (I) that indexes the dictionary, and a respective lookup coefficient tensor (C), and in which applying the convolutional neural network includes: convolving the channel weight vectors of the dictionary (D) with an input tensor based on the image to obtain an input dictionary (S), and combining entries of the input dictionary (S) that are indexed with indices from the respective lookup index tensor (I) and multiplied with corresponding coefficients from the respective lookup coefficient tensor (C); and storing, displaying, or transmitting data based on the output of the convolutional neural network.

Abstract: In order to detect an object by discriminating the object in accordance with the state of the object, a detection device comprises: an identifying unit configured to identify whether a human included in a plurality of frame images of a moving image is a pedestrian; a determining unit configured to determine, for notifying, that the human is the pedestrian if the identifying unit identifies that the number of times the human is the pedestrian satisfies a predetermined condition; and a notifying unit configured to notify intrusion if the human determined as the pedestrian by the determining unit intrudes into an area preset for the moving image.

Abstract: A method for processing angiography image data by using an imaging catheter path that is directly detected from the angiography data as a co-registration path or using detected marker locations from the angiography data to generate a co-registration path. If the acquired angiography data includes synchronized cardiac phase signals and a predetermined quantity of angiography image frames not including contrast media, then a directly detected imaging catheter path is used as the co-registration path. Otherwise the co-registration path is determined based upon detected marker locations from the angiography image data.

Abstract: A method, apparatus, electronic device, and a system for selecting a to-be-masked region in a video are disclosed. The method includes: obtaining a video to be detected; determining specified target sets in the video through a preset target detection algorithm, wherein each specified target set is a set of pixels of one specified target in video frames of the video; determining whether a specified target corresponding to each specified target set is a sensitive target respectively through a preset recognition algorithm; and using a specified target set as to-be-masked regions in the video when a specified target corresponding to this specified target set is determined as a sensitive target. In this method, specified targets are first determined, and then it is determined whether the specified targets are sensitive targets.

Abstract: A presence detection system includes a camera that captures an image; an appearance-based detector that detects appearance of an object according to the image, thereby outputting an appearance-detection signal that is asserted when the appearance is detected; a proximity detector that detects motion of the object on a predetermined detection window on the image, thereby outputting a proximity-detection signal that is asserted when the motion is detected; and a fusion unit that receives the appearance-detection signal and the proximity-detection signal, thereby outputting an object-presence signal that is asserted when the appearance-detection signal or the proximity-detection signal is asserted.

Abstract: Disclosed are devices, systems, methods, techniques, and computer program products for estimating a Region Of Interest (ROI) corresponding to a plurality of content streams. A method may include receiving a plurality of sensor data associated with a plurality of mobile devices. The plurality of sensor data corresponds to a plurality of content streams captured by the plurality of mobile devices. Further, each of the plurality of mobile devices may include at least one recorder configured to capture a corresponding content stream. Further, a sensor data associated with a mobile device may include a location data and an orientation data of the mobile device during capturing of a content stream. The method may further include, analyzing, by the system, the plurality of sensor data and determining, by the system, a ROI based on at least the plurality of sensor data, thereby generating an estimation of the ROI.

Abstract: A medical image processing apparatus for allocating at least two medical imaging processes to a plurality of assignable processing resources is provided. The plurality of assignable processing resources is allocated by the medical image processing apparatus based on resource information of the plurality of assignable processing resources. The medical image processing apparatus includes circuitry configured to acquire medical image processing content, from medical equipment, to be processed according to the at least two medical imaging processes prior to display on a display device connected to a surgical operating room network. The circuitry is configured to acquire the resource information of the plurality of assignable processing resources, and allocate each of the at least two medical imaging processes to a different one of the plurality of assignable processing resources based on the resource information of the plurality processing resources and the medical image processing content.

Abstract: Techniques detecting usage of copyrighted video content using object recognition are provided. In one example, a computer-implemented method comprises determining, by a system operatively coupled to a processor, digest information for a video, wherein the digest information comprises objects appearing in the video and respective times at which the objects appear in the video. The method further comprises comparing, by the system, the digest information with reference digest information for reference videos, wherein the reference digest information identifies reference objects appearing in the reference videos and respective reference times at which the reference objects appear in the reference videos. The method further comprises determining, by the system, whether the video comprises content included in one or more of the reference videos based on a degree of similarity between the digest information and reference digest information associated with one or more of the reference videos.

Abstract: There is provided a moving robot including a light projector, an image sensor and a processing unit. The light projector projects a vertical light segment toward a moving direction. The image sensor captures, toward the moving direction, an image frame containing a light segment image associated with the vertical light segment. The processing unit calculates a step distance and a segment feature according to the image frame, outputs a flag signal according to the segment feature to indicate whether the calculated step distance is confident or not, and perform a pixel interpolation in calculating the step distance to improve the identification accuracy.

Abstract: Systems, and method and computer readable media that store instructions for motion based object detection. The method may include receiving or generating a video stream that comprises a sequence of images; generating image signatures of the images; wherein each image is associated with an image signature that comprises identifiers; wherein each identifier identifiers a region of interest within the image; generating movement information indicative of movements of the regions of interest within consecutive images of the sequence of images; searching, based on the movement information, for a first group of regions of interest that follow a first movement; wherein different first regions of interest are associated with different parts of an object; and linking between first identifiers that identify the first group of regions of interest.

Abstract: Embodiments of the present invention provide a system and a method for posture estimation, comprising: capturing a video sequence by a mobile terminal, the video sequence including a plurality of frames; extracting a plurality of convolutional features of a subject in the video sequence; estimating pose of a subject from the video sequence; constructing a model of the plurality of features from the estimated pose; estimating the posture of the subject from the plurality of features. The system also determines a key posture sequence from unconstrained video, where a mobile terminal captures a video sequence; a tracker estimates a reference subject; a posture estimator estimates a posture of the subject from plurality of features; a posture analytic unit assesses the estimated posture; a posture reconstructing unit reconstructs the posture into three dimensional space; an evaluating unit determines the estimated posture that should be added to key posture sequence.

Abstract: A robotic system includes a processor that is programmed to determine and cause work site measurements for user specified points in the work site to be graphically displayed in order to provide geometrically appropriate tool selection assistance to the user. The processor is also programmed to determine an optimal one of a plurality of tools of varying geometries for use at the work site and to cause graphical representations of at least the optimal tool to be displayed along with the work site measurements.

Abstract: Systems and methods to label images for inclusion into a machine learning image recognition training data set. A coded labeling definition is defined. A set of digital images is collected that includes images of different types of piece of equipment. A digital image is presented that includes an image of a particular piece of equipment. A received numeric code corresponding to the particular piece of equipment within the present image is received. The received numeric code is associated with the present image and the present image is stored in association with the received numeric code into a machine learning image recognition data set. A machine learning image recognition is trained based on the machine learning image recognition data set to automatically associate unlabeled images with the respective numeric code that corresponds to the particular piece of equipment in the unlabeled image.

Abstract: A method for monocular depth/pose estimation in a camera agnostic network is described. The method includes training a monocular depth model and a monocular pose model to learn monocular depth estimation and monocular pose estimation based on a target image and context images from monocular video captured by the camera agnostic network. The method also includes lifting 3D points from image pixels of the target image according to the context images. The method further includes projecting the lifted 3D points onto an image plane according to a predicted ray vector based on the monocular depth model, the monocular pose model, and a camera center of the camera agnostic network. The method also includes predicting a warped target image from a predicted depth map of the monocular depth model, a ray surface of the predicted ray vector, and a projection of the lifted 3D points according to the camera agnostic network.

Abstract: Systems, methods and computer program products are provided for reading and scoring ultrasound and/or optoacoustic (US/OA) images that include at least one of OA images or US images acquired in connection with an examination for a region of interest (ROI). The system displays a first image that has an interior ROI outline separating an internal zone from a boundary zone. In some aspects, feature scores are obtained in connection with at least the boundary zone and peripheral zone of the first image and the feature scores are applied to a classification model to obtain at least one of a prognostic result or predictive result indicative of a trait of the lesion. In accordance with some aspects, an order in which feature scores are entered is automatically managed to obtain for the at least one of the peripheral zone or the boundary zone before the feature score is obtained for the internal zone.

Abstract: Technology for constructing a symbol template for use by a 1D barcode decoder. One method includes obtaining a first symbol pattern representing a first symbol, the first symbol pattern comprising a plurality of elements, with each element corresponding to a 1D barcode space or bar, in which the first symbol represents one unique 1D bar (b) and space (s) pattern corresponding to a symbol or digit: 0, 1, 2, 3, 4, 5, 6, 7, 8 or 9; obtaining prefix elements, the prefix elements comprising a subset of second symbol elements from a second symbol pattern which precedes the first symbol from within a scanline of image data, the subset being less than a total number of elements within the second symbol elements; extending the first symbol pattern with the prefix elements to yield an overlapping symbol template for use in correlation-based barcode decoding. Other technology provided as well.

Abstract: The purpose of the present invention is to improve the accuracy of line-of-sight estimation based on an image. A line-of-sight estimation device 100 comprises: an estimation unit 110 which estimates lines of sight from a face included in a face image, using a plurality of estimators; and a determination unit 120 which determines a line of sight for the face on the basis of first condition information including conditions relating to the capture of the face image, and a plurality of sets of second condition information, each including conditions associated with one of the plurality of estimators, and on the basis of a plurality of estimated lines of sight.

Abstract: An image is reconstructed an image from projection data by an image reconstruction computing entity. Projection data captured by an imaging device is received. An image is reconstructed and/or generated based at least in part on the projection data using a forward and back projection technique employing a convolutional spline process. The image is provided such that a user computing entity receives the image. The user computing entity is configured to display the image via a user interface thereof.

Abstract: The purpose of the present invention is to improve the accuracy of line-of-sight estimation based on an image. A line-of-sight estimation device 100 comprises: an estimation unit 110 which estimates lines of sight from a face included in a face image, using a plurality of estimators; and a determination unit 120 which determines a line of sight for the face on the basis of first condition information including conditions relating to the capture of the face image, and a plurality of sets of second condition information, each including conditions associated with one of the plurality of estimators, and on the basis of a plurality of estimated lines of sight.

Abstract: The purpose of the present invention is to improve the accuracy of line-of-sight estimation based on an image. A line-of-sight estimation device 100 comprises: an estimation unit 110 which estimates lines of sight from a face included in a face image, using a plurality of estimators; and a determination unit 120 which determines a line of sight for the face on the basis of first condition information including conditions relating to the capture of the face image, and a plurality of sets of second condition information, each including conditions associated with one of the plurality of estimators, and on the basis of a plurality of estimated lines of sight.