Abstract: The method comprising determining a set of coordinates each for two or more appearances of a target subject within a sequence of images, the set of coordinates of the two or more appearances of the target subject defining a first path; determining a set of coordinates each for two or more appearances of a related subject within a sequence of images, the related subject relating to the target subject, the set of coordinates of the two or more appearances of the related subject defining a second path; determining one or more minimum distances between the first path and the second path so as to determine at least a region of interest; determining a timestamp of a first appearance and a timestamp of a last appearance of the target subject; and determining a timestamp of a first appearance and a timestamp of a last appearance of the related subject.

Abstract: The embodiments of the present disclosure disclose image defogging method. A specific mode of carrying out the method includes the steps to: obtain a polarized fog-free scene image set, generate a first quintuple based on each polarized fog-free scene image, and obtain a first quintuple set; extract pixel coordinates meeting a preset condition from the significant polarization image included in each first quintuple, to obtain a pixel coordinate group to generate a second quintuple; generate a simulated polarization foggy scene image sequence based on each second quintuple, to obtain a simulated polarization foggy scene data set; design a polarization state attention neural network; obtain a polarized foggy scene image, and input the polarized foggy scene image into the polarization state attention neural network to obtain a polarized defogging image. This embodiment can still improve the accuracy of image defogging with poor lighting conditions.

Abstract: Methods, systems, and apparatus for high-fidelity vision tasks using deep neural networks are disclosed. An example apparatus includes a feature extractor to extract low-level features and edge-enhanced features of an input image processed using a convolutional neural network, an eidetic memory block generator to generate an eidetic memory block using the extracted low-level features or the extracted edge-enhanced features, and an interactive segmentation network to perform image segmentation using the eidetic memory block, the eidetic memory block used to propagate domain-persistent features through the segmentation network.

Abstract: The present application relates to a method for recognising at least one object in a three-dimensional scene, the method including, in an electronic processing device: determining a plurality of two-dimensional images of the scene, the images at least partially including the at least one object; determining a plurality of two-dimensional segmentations of the at least one object, the two-dimensional segmentations corresponding to the two dimensional images; generating a three-dimensional representation of the scene using the images; generating a mapping indicative of a correspondence between the images and the representation; and using the mapping to map the plurality of segmentations to the three dimensional representation, to thereby recognise the at least one object in the scene.

Abstract: The method comprising determining a set of coordinates each for two or more appearances of a target subject within a sequence of images, the set of coordinates of the two or more appearances of the target subject defining a first path; determining a set of coordinates each for two or more appearances of a related subject within a sequence of images, the related subject relating to the target subject, the set of coordinates of the two or more appearances of the related subject defining a second path; determining one or more minimum distances between the first path and the second path so as to determine at least a region of interest; determining a timestamp of a first appearance and a timestamp of a last appearance of the target subject; and determining a timestamp of a first appearance and a timestamp of a last appearance of the related subject.

Abstract: Various embodiments are generally directed to techniques for image segmentation utilizing context, such as with a machine learning (ML) model that injects context into various training stages. Many embodiments utilize one or more of an encoder-decoder model topology and select criteria and parameters in hyper-parameter optimization (HPO) to conduct the best model neural architecture search (NAS). Some embodiments are particularly directed to resizing context frames to a resolution that corresponds with a particular stage of decoding. In several embodiments, the context frames are concatenated with one or more of data from a previous decoding stage and data from a corresponding encoding stage prior to being provided as input to a next decoding stage.

Abstract: Provided are template mark detection method and template position correction method based on single camera, including: performing image collection on mark on template by single camera, and obtaining binary image after preprocessing; performing corner detection of jagged edges on binary image to obtain corner set of jagged edges; performing edge detection and line detection sequentially on binary image to obtain set of edge line segments from coarse detection; traversing such set, and judging and retaining collinear line segments, to obtain set of collinear line segments from coarse detection; traversing corner set of jagged edges, for point-line collinearity judgment with line segments in set of collinear line segments from coarse detection, to obtain set of point-line from fine detection; and performing linear fitting on set of point-line from fine detection, and calculating an inclination angle of each straight line through an arctan function, thus completing detection of mark.

Abstract: A monitoring system based on a digital converter station includes a first monitoring terminal deployed at a first-level monitoring side, a second monitoring terminal deployed at a second-level monitoring side, and a third monitoring terminal deployed at a third-level monitoring side; the monitoring terminal at each level includes a communication module, a human-machine interaction module, a device status monitoring module, a device alarm management module, a video fusion processing module, and a data transmission adjustment and control module.

Abstract: A method for inspecting a vehicle on a site with a management system, including determining whether at least one vehicle on the site intends to make a departure from a stoppage place, generating a control command by the management system, and transmitting the control command to the inspection vehicle such that the inspection vehicle moves along an inspection path to the stoppage place. The method further includes detecting an environment around the inspection vehicle at the stoppage place by the sensor system while a detection zone of the at least one sensor is oriented at least in some regions at a subregion beneath the vehicle and/or at the vehicle. The method further includes determining, in dependence on the sensor signals generated by the sensor system, whether there is an impediment by which the departure of the vehicle is impeded, in order to carry out a pre-departure check.

Abstract: An example method comprises receiving a new observation characterizing at least one parameter of an entity; inputting the new observation to a deep neural network having hidden layers; obtaining a second set of intermediate output values that are output from at least one of the hidden layers by inputting the received new observation to the deep neural network; mapping the second set of intermediate output values to a second set of projected values; determining whether or not the received new observation is an outlier with respect to the training dataset based on the latent variable model and the second set of projected values, calculating a prediction for the new observation; and determining a result indicative of the occurrence of at least one anomaly in the entity based on the prediction and the determination whether or not the new observation is an outlier.

Abstract: Disclosed is a defect inspection device. The defect inspection device may include a lighting system designed for transmitting a lighting pattern having different illuminances for each area on a surface of an inspection object; a photographing unit for obtaining an image data of the inspection object; one or more processors for processing the image data; and a memory for storing a deep learning-based model. In addition, the one or more processors are adapted to control, the lighting system to transmit a lighting pattern having a different illuminance for each area on a surface of an inspection object, input, an image data obtained by the photographing unit into the deep learning-based model, wherein the image data includes a rapid change of illuminance in at least a part of the object surface; and determine, a defect on a surface of the inspection object using the deep learning-based model.

Abstract: Disclosed in various embodiments of the present invention are a method and a device for measuring a user's biometric information in an electronic device and providing information related to the biometric information. An electronic device according to various embodiments of the present invention comprises a sensor module, a camera module, a display device, and a processor, wherein the processor can be configured to: execute an application; acquire a user's first biometric information on the basis of the sensor module while the operation relating to the application is performed; estimate a user's health information at least one the basis of the first biometric information, and link the health information with the operation relating to the application so as to display same through the display device. Various embodiments are possible.

Abstract: A method is provided for indoor multipath ghosts recognition for a multiple-input-multiple-output radar with collocated antennas. The method includes the step of generating a two-dimensional Range-Doppler map for N number of consecutive radar data frames. The method further includes the step of applying a temporal clustering algorithm to the N number of consecutive radar data frames. Moreover, the method includes the step of applying a linear pattern extraction algorithm on the two-dimensional Range-Doppler map. In this context, the two-dimensional Range-Doppler map comprises detections from at least one target, at least one first-order ghost, and at least one second-order ghost with respect to one wall reflector.

Abstract: In one embodiment, a method includes accessing a stream of F video frames, where each of the F video frames includes N patches that are non-overlapping, generating an initial embedding vector for each of the N×F patches in the F video frames, generating a classification embedding by processing the generated N×F initial embedding vectors using a self-attention-based machine-learning model that computes a temporal attention and a spatial attention for each of the N×F patches, and determining a class of the stream of video frames based on the generated classification embedding.

Abstract: A method and apparatus for generating an image of vehicle surroundings are disclosed, including: capturing vehicle surroundings by vehicle cameras arranged on a vehicle body of a vehicle, and generating camera images by the cameras. The camera images of adjacent cameras have overlapping image regions), generating a virtual representation of the surroundings in a virtual three-dimensional space, during which the camera images are projected onto a virtual projection surface in the space. A non-stationary virtual camera in the virtual space determines a position and/or orientation thereof. A first selection region is placed on the surface in a first overlapping image region depending on a virtual camera field of vision, at least one image parameter of a first vehicle camera in the first selection region is calculated, and at least one second vehicle image parameter is adjusted to the at least one first vehicle image parameter in the first selection region.

Abstract: A method including generating a captured facial object and a captured pose from a captured image. The method also includes obtaining a base facial object and a base pose from a base image. The method also includes generating base pose angles using the captured pose, and captured pose angles using the captured pose. The method also includes obtaining selected base images using the base pose angles and the base facial object. The method also includes generating selected captured images using the captured pose angles and the captured facial object. The method also includes comparing the selected base images to the selected captured images to establish a comparison. The method also includes outputting a match output using the comparison.

Abstract: Quantifying plant infestation is performed by estimating the number of biological objects (132) on parts (122) of a plant (112). A computer (202) receives a plant-image (412) taken from a particular plant (112). The computer (202) uses a first convolutional neural network (262/272) to derive a part-image (422) that shows a part of the plant. The computer (202) splits the part-image into tiles and uses a second network to process the tiles to density maps. The computer (202) combines the density maps to a combined density map in the dimension of the part-image and integrates the pixel values to an estimate number of objects for the part. Object classes (132(1), 132(2)) can be differentiated to fine-tune the quantification to identify class-specific countermeasures.

Abstract: A stress analysis device includes: an imaging element that obtains temperature images over a same time range for a same region of an object; a feature point extractor that extracts a feature point in each of the temperature images; a projection transformer that performs projective transformation on each of the temperature images to align the feature point in the temperature images, and aligns the temperature images with respect to a temperature image being a reference; a pixel rearranger that rearranges a pixel array of each of the temperature images subjected to the projective transformation with respect to a pixel array of the temperature image that is the reference; a stress converter that obtains a stress image by multiplying each of the temperature images after pixel rearrangement by a stress conversion coefficient; and an additional averaging part that obtains an additional averaging stress image by adding and averaging the stress images.

Abstract: A method for identifying charcoal products includes capturing, by a camera, an image of a sample of charcoal. The method also includes analyzing, by a processor in communication with the camera, the image of the sample to identify characteristics of the sample. The method also includes comparing, by the processor, the identified characteristics to a plurality of signatures to determine a match between the identified characteristics and known characteristics associated with each signature in the plurality of signatures. The method further includes generating, by the processor, an output based on the comparison.

Abstract: The invention provides an image encoding apparatus operable to encode data obtained by an image capturing sensor in which a filter for detecting a fourth color is periodically arranged in addition to filters of three primary colors, where the apparatus comprises a generating unit configured to generate data that approximates the fourth color using data of at least two colors among three colors, which represent three primary colors, obtained by the image capturing sensor and generate difference data that represents a difference between the generated data and data of the fourth color; and an encoding unit configured to encode data of the three colors which represent three primary colors and the difference data.