Abstract: Disclosed are a target tracking method and apparatus based on a path. The target tracking method may include displaying an image that is received through a camera included in a target tracking apparatus on a display included in the target tracking apparatus, setting two or more points based on the image displayed on the display, and tracking a path that is determined based on the two or more points by controlling a rotation of the camera through a driving unit included in the target tracking apparatus.

Abstract: A method of crowdsourcing annotation for medical image data based on image comparison, includes developing an annotation task; acquiring and verifying medical images according to requirements of the annotation task, and performing statistics on the medical images that meet verification standards to obtain an amount of image data acquired; archiving the medical images that meet the verification standards to form a medical image database; setting up annotation rules, and generating annotation schemes and annotation samples according to the requirements of the annotation task and annotation rules; distributing the medical images in the medical image database to multiple crowdsourcing annotation sides, so that each medical image being annotated by the multiple crowdsourcing annotation sides; annotating, by the crowdsourcing annotation sides, the distributed medical images according to the annotation schemes and the annotation examples to form the annotated images; and merging multiple annotated images correspondi

Abstract: A real-time, intra-uterine operating field image is displayed to an operator. A processor identifies anatomical features and marks anatomical boundaries on the image with the aid of a trained classifier model. Labels for the identified anatomical features and the anatomical boundaries are displayed on the image. The trained classifier model is trained based on a set of prior surgical field images that have anatomical features and anatomical boundaries therein identified and marked, respectively, by previous operators. The trained classifier model is typically a convolutional deep neural network. The operator is allowed to modify the anatomical feature labels and anatomical boundaries, as well as update the set of training images with the modified image. The labeling of the anatomical features and establishment of the anatomical features facilitate performing treatments in the uterus, such as the ablation of uterine fibroids.

Abstract: A system for selecting motion models for aligning scene content captured by different image sensors, is configurable to (i) access a first image captured by a first image sensor and a second image captured by a second image sensor; (ii) access a set of motion models; (iii) define a reference patch within the second image; (iv) generate a respective match patch for each motion model of the set of motion models; (v) determine a similarity between each respective match patch and the reference patch within the second image; (vi) select a final motion model from the set of motion models based upon the similarity between each respective match patch and the reference patch within the second image; and (vii) utilize the final motion model to generate an output image for display to a user.

Abstract: Systems, components, methods, and algorithms for generating template information are described. a computer-implemented method includes receiving microscope image data, detecting a feature in the microscope image data, receiving feature template data describing a grouping of features including the feature, detecting the grouping in the microscope image data based at least in part on the feature template data, generating template information describing the grouping, and outputting the template information.

Abstract: In various examples, circadian rhythm-based data augmentation for drowsiness detection systems and applications are provided. Embodiments described herein may produce an estimated circadian rhythm for a test subject and/or vehicle driver or other machine operator or occupant, and use the pattern of that circadian rhythm to correct, confirm, calibrate, or otherwise augment drowsiness assessments derived from video image data. The position of a person in the context of their process C circadian cycle may be used as indication of their level of drowsiness. An estimated process C circadian cycle may be used to generate more accurate ground truth training data for training machine learning models, and may be used by real-time, in-vehicle drowsiness detection systems that infer driver drowsiness levels based on captured images. In various embodiments, a circadian rhythm drowsiness estimate may be used to correct, calibrate, augment, and/or replace a drowsiness score predicted by a machine learning model.

Abstract: A system for correcting an alignment of positions of points affiliated with an object, in images of a location, that has one or more of a linear feature or a planar feature can include a processor and a memory. The memory can store an alignment module and a communications module. The alignment module can include instructions to: (1) identify, within data affiliated with the images, the positions of the points affiliated with the object that has the one or more of the linear feature or the planar feature and (2) correct, in a manner that recognizes that the object has the one or more of the linear feature or the planar feature, the alignment of the positions to produce a digital map of the location. The communications module can include instructions to transmit the digital map to a vehicle to be used to control a movement of the vehicle.

Abstract: The present disclosure provides a method and an apparatus for estimating a body part position of a person, the method comprising: detecting, by a processor, a plurality of body part positions of the person from an image in which the person is detected, wherein each of the plurality of body part positions corresponds to a body part of the person; and estimating, by the processor, a body part position of the person based on the detected plurality of body part positions.

Abstract: A method and a system for detecting railway foreign object intrusion, and a device and a medium can include acquiring a first key frame of railway image of a current scenario as a template image, and an Nth frame of railway image as a to-be-detected image; inputting the template image and the to-be-detected image of the current scenario into a foreign object intrusion prediction model to obtain a foreign object feature map of the current scenario, so as to determine whether there is a different target between the to-be-detected image and the template image of the current scenario; and filtering, by using a moving object filter model, a moving object from the different target of the to-be-detected image of the current scenario in response to existing different target, to obtain a true foreign object in the to-be-detected image of the current scenario.

Abstract: A system and a method are described for extracting a target area of an aviation document. The system and method include inputting the aviation document to an image segmentation model. The aviation document includes a target area of interest that is shown to scale. The image segmentation model generates a first output that provides a predicted location of the target AOI relative to the aviation document. The aviation document is input to an edge/contour detection algorithm that generates a second output which identifies a plurality of candidate polygons based on content in the aviation document. The first output is analyzed with the second output to determine a selected polygon of the candidate polygons in the second output that corresponds with the predicted location of the target AOI in the first output. Pixel coordinate values of the selected polygon in the aviation document are determined.

Abstract: A method for identifying a plurality of blades of an engine. The method can include obtaining and recording a reference view captured by a vision system of an initial, or first, blade of a plurality of blades of the engine at a first blade position, such as an inspection position. The positions of other blades that are to be subsequently identified and/or cataloged can be evaluated by a controller from information captured by the vision system and the reference view to determine if the subsequent blade is at the first blade position. If a blade is determined to not be at the first blade position, the controller can generate an error signal indicative of an amount an actuator is to be operated to displace the blade to the first blade position. The controller can compare information captured from the image of a blade with stored data to identify the blade.

Abstract: Systems and methods for vegetation detection from aerial photogrammetric multispectral data. The method includes: detecting apexes in a height model using Local Maxima (LM) detection; detecting vegetation as detected apexes; performing orthorectification to derive an orthomosaic; generating a fractional map of a vegetation class by applying a Fuzzy classifier on the orthomosaic using the detected vegetation to define the class; generating a binary ridge map using the height model to identify ridges; combining the binary ridge map and the fractional map of the vegetation class to generate a ridge integrated fractional map; performing delineation of individual vegetation on the ridge integrated fractional map on a vegetation class using an active contour algorithm; and outputting the delineated vegetation.

Abstract: Various of the disclosed embodiments relate to systems and methods for processing surgical data to facilitate further downstream operations. For example, some embodiments may include machine learning systems trained to recognize whether video from surgical visualization tools, such as endoscopes, depicts a field of view inside or outside the patient body. The system may excise or whiteout frames of video appearing outside the patient so as to remove potentially compromising personal information, such as the identities of members of the surgical team, the patients identity, configurations of the surgical theater, etc. Appropriate removal of such non-surgical data may facilitate downstream processing, e.g., by complying with regulatory requirements as well as by removing extraneous data potentially inimical to further downstream processing, such as training a downstream classifier.

Abstract: An image processing apparatus includes a processor, and a memory storing a plurality of images obtained by capturing images of a building and a three-dimensional model of the building in which members constituting the building are specified, the plurality of images and the members being stored in association with each other. The processor is configured to perform an extraction process to extract defect information of the building on the basis of the plurality of images, a selection process to select an image corresponding to a designated member from among the plurality of images in accordance with a designated criterion, and an output process to output the designated member, the selected image, and the defect information in association with each other.

Abstract: A method can include identifying handwritten elements from a document image. The method can include generating embeddings for the handwritten elements, and the embeddings may represent features extracted from the handwritten elements. The method can include clustering the embeddings into clusters. The method can include processing the document image to identify areas within the document image designated for placement of a signature. The method can include, based upon locations of the areas within the document image, tagging each cluster as a signature cluster or as a non-signature cluster. The method can include outputting indications that a handwritten element is a signature element or is a non-signature element.

Abstract: Techniques are disclosed for generating an altered version of an original version of media using a generative model. A system detects, at a computing device, a trigger event indicating a user of the computing device has requested to view media at a user interface of the computing device. In response to detecting the trigger event, the system: retrieves, an original version of the media from a backend server and identifies, using an object detection model, a bounding region of content within the original version, generates, using a generative model, an altered version of the media based on the bounding region, and transmits to the computing device, the altered version for display at the computing device in place of the original version, where the generative model generates the altered version of the media based on historical transaction information and user information of the user of the computing device.

Abstract: A method for controlling an artificial intelligence (AI) device can include receiving, via a processor, a base dataset, and receiving, via the processor, an image that is unlabeled. Also, the method can include inputting the image to at least one of a caption-based pipeline and a data programming pipeline to generate a labeled image training data sample including triplet information, the caption-based pipeline including a matching model configured to receive textual scene graph information for the image and bounding box information for the image, and the data programming pipeline including a feature extraction model configured to output three vectors and a label generator configured to receive the three vectors. The method can further include in response to the predicate matching a same predicate in a tail distribution of the base dataset, merging the labeled image training data sample with the base dataset to generate an enhanced dataset.

Abstract: A method for feature fusion model training and sample retrieval includes: inputting training sample into an initial feature fusion model to obtain a training semantic feature and a training global feature, performing classification and recognition based on the training semantic feature to obtain an initial training category, splicing the training semantic feature and the training global feature to obtain a spliced training feature, performing autocorrelation feature calculation based on the spliced training feature to obtain an autocorrelation feature, performing self-attention weight calculation based on the autocorrelation feature to obtain a self-attention weight, and adjusting the spliced training feature through the self-attention weight to obtain a fused training feature; and updating the initial feature fusion model based on the training global feature, the training semantic feature, the fused training feature, the initial training category, and a training sample category label, and performing a loop i

Abstract: In one embodiment, an apparatus includes accessing an image, captured by a first camera, of an object that includes a size-invariant feature and a size-variant feature and determining a size of the size-invariant feature in the image. The method further includes determining, based on the determined size of the size-invariant feature in the image, a distance between the camera and the object and determining a size of the size-variant feature in the image. The method further includes determining, based on the determined distance between the camera and the object and based on the determined size of the size-variant feature in the image, an actual size of the size-variant feature.

Abstract: A gesture recognition device that recognizes, based on an image obtained by imaging a person to be imaged with a digital camera having an imaging range changing mechanism for changing an imaging range, a gesture represented by a hand of the person to be imaged includes at least one processor. The at least one processor controls, in a case where a position of a face of the person to be imaged in the image deviates from a predetermined set position, an operation of the imaging range changing mechanism to cause the digital camera to capture a proper face position image with the position of the face as the set position, and determines whether the hand detected from the proper face position image is a right hand or a left hand based on a positional relationship between a position of the detected hand and the set position.