Abstract: Disclosed are a deep neural network lightweight device based on batch normalization, and a method thereof. The deep neural network lightweight device based on batch normalization includes a memory that stores at least one data and at least one processor that executes a network lightweight module. When executing the network lightweight module, the processor performs learning on an input neural network based on sparsity regularization to adaptively determine at least one parameter of the sparsity regularization, performs pruning on the learning result, and performs fine tuning on the pruning result.

Abstract: The present invention relates to an apparatus and method for drawing, the method comprising: inputting a drawing image; recognizing a component in the input drawing image; inferring a structure of an object based on the recognized component; and drawing the inferred structure of the object.

Abstract: Provided are a method and system for training a dynamic deep neural network. The method for training a dynamic deep neural network includes receiving an output of a last layer of the deep neural network and outputting a first loss, receiving an output of a routing module according to an input class of the deep neural network and outputting a second loss, calculating a third loss based on the first loss and the second loss, and updating a weight of the deep neural network by using the third loss.

Abstract: An apparatus and method for searching for a building on the basis of an image and a method of constructing a building search database (DB) for image-based building search. The method includes constructing a building search DB, receiving a query image from a user terminal, detecting a region to which a building belongs in the query image, extracting features of the region detected in the query image, and searching the building search DB for a building matching the extracted features. Therefore, building search performance can be improved.

Abstract: A data learning device in a deep learning network characterized by a high image resolution and a thin channel at an input stage and an output stage and a low image resolution and a thick channel in an intermediate deep layer includes a feature information extraction unit configured to extract global feature information considering an association between all elements of data when generating an initial estimate in the deep layer; a direct channel-to-image conversion unit configured to generate expanded data having the same resolution as a final output from the generated initial estimate of the global feature information or intermediate outputs sequentially generated in subsequent layers; and a comparison and learning unit configured to calculate a difference between the expanded data generated by the direct channel-to-image conversion unit and a prepared ground truth value and update network parameters such that the difference is decreased.

Abstract: A fast object detection method and a fast object detection apparatus using an artificial neural network. The fast object detection method includes obtaining an input image; inputting the obtained input image into an object detection neural network using a plurality of preset bounding boxes; and detecting an object included in the input image by acquiring output data of the object detection neural network.

Abstract: An apparatus and method for searching a neural network architecture may be disclosed. The apparatus may include an architecture searcher and an architecture evaluator. The architecture searcher may search for a topology between nodes included in a basic cell of a network, search for an operation to be applied between the nodes after searching for the topology, and determine the basic cell. The architecture evaluator may evaluate performance of the determined basic cell.

Abstract: A fast object detection method and a fast object detection apparatus using an artificial neural network. The fast object detection method includes obtaining an input image; inputting the obtained input image into an object detection neural network using a plurality of preset bounding boxes; and detecting an object included in the input image by acquiring output data of the object detection neural network.

Abstract: A machine learning method for learning how to form bounding boxes, performed by a machine learning apparatus, includes extracting learning images including a target object among a plurality of learning images included in a learning database, generating additional learning images in which the target object is rotated from the learning images including the target object, and updating the learning database using the additional learning images.

Abstract: An apparatus and method for searching for a building on the basis of an image and a method of constructing a building search database (DB) for image-based building search. The method includes constructing a building search DB, receiving a query image from a user terminal, detecting a region to which a building belongs in the query image, extracting features of the region detected in the query image, and searching the building search DB for a building matching the extracted features. Therefore, building search performance can be improved.

Abstract: An omni-channel management method includes receiving a distribution channel information and a product purchase information from each of a plurality of distribution channels, and integrating the received distribution channel information and the product purchase information; applying an image identification technology and a voice identification technology to retrieve and integrate information related to a product; receiving and integrating product information from a manufacturer, a seller, or a web site; verifying consistency of the integrated product information by checking whether detailed information included in the integrated product information is related to a same product; and providing the consistency-secured product information to a consumer and a plurality of distribution channels.

Abstract: A machine learning method for learning how to form bounding boxes, performed by a machine learning apparatus, includes extracting learning images including a target object among a plurality of learning images included in a learning database, generating additional learning images in which the target object is rotated from the learning images including the target object, and updating the learning database using the additional learning images.

Abstract: An apparatus for object detection according to an example includes a level image generating unit configured to generate a plurality of level images with reference to a target image; a feature vector extracting unit configured to extract a feature vector from each level image; a codeword generating unit configured to generate a codeword by clustering the feature vector for each level image; a histogram generating unit configured to generate a histogram corresponding to the codeword; and a classifier configured to generate object recognition information of the target image based on the histogram.

Abstract: Disclosed herein is an apparatus and method of searching for a wanted vehicle, capable of interoperating with black boxes mounted in vehicles of unspecified individuals, recognizing and searching for registration numbers of vehicles in proximity of each black box in real time, and identifying a location of the wanted vehicle in real time using information about locations of the searched vehicles. The method includes requesting, by an apparatus for searching for a wanted vehicle, a black box installed in at least one vehicle to search for a registration number of the wanted vehicle, and receiving a response corresponding to the request, and acquiring information about the wanted vehicle and a location of the wanted vehicle corresponding to the response using the black box for recognizing a vehicle registration number or the black box for detecting a vehicle registration number region.

Abstract: The present invention relates to a method of tracking an object in a multiple cameras environment and the method includes generating first feature information of the object from an image input from a first camera; detecting a second camera in which identification information for the object is recognized when the object moves out of a view angle of the first camera, and comparing second feature information of the object generated from an image input from the second camera with the first feature information to track the object from the image input from the second camera. According to the present invention, the object is tracked based on an image in one camera image and if the object moves out of the camera, the identification information of the terminal which is possessed by the object is recognized to hand over the camera to continuously track the same object.

Abstract: A human tracking method using a color histogram is disclosed. The human tracking method using the color histogram according to the present invention can more adaptively perform human tracking using different target color histograms according to the human poses, instead of applying only one target color histogram to the tracking process of one person, such that the accuracy of human tracking can be increased. The human tracking method includes performing color space conversion of input video data; calculating a state equation of a particle based on the color-space conversion data; calculating the state equation, and calculating human pose-adaptive observation likelihood; resampling the particle using the observation likelihood, and estimating a state value of the human; and updating a target color histogram.

Abstract: Disclosed herein is an apparatus and method of searching for a wanted vehicle, capable of interoperating with black boxes mounted in vehicles of unspecified individuals, recognizing and searching for registration numbers of vehicles in proximity of each black box in real time, and identifying a location of the wanted vehicle in real time using information about locations of the searched vehicles. The method includes requesting, by an apparatus for searching for a wanted vehicle, a black box installed in at least one vehicle to search for a registration number of the wanted vehicle, and receiving a response corresponding to the request, and acquiring information about the wanted vehicle and a location of the wanted vehicle corresponding to the response using the black box for recognizing a vehicle registration number or the black box for detecting a vehicle registration number region.

Abstract: Provided is a trajectory modeling apparatus and method based on trajectory transformation which models a trajectory and compares a trajectory of an object and the modeled trajectory. The trajectory modeling apparatus includes an image input unit configured to receive an input image, an object trajectory generating unit configured to trace an object included in the input image to generate a trajectory of the object, a trajectory model generating unit configured to generate a trajectory model according to a directionality of the trajectory of the object by using the trajectory of the object, and a trajectory analyzing unit configured to analyze a trajectory of a target included in a test image by using the trajectory model to determine whether a behavior of the target is normal, based on the target trajectory analysis result.

Abstract: Disclosed is a learning method using extracted data features for simplifying a learning process or improving accuracy of estimation. The learning method includes dividing input learning data into two groups based on a predetermined reference, extracting data features for distinguishing the two divided groups, and performing learning using the extracted data features.

Abstract: A human tracking method using a color histogram is disclosed. The human tracking method using the color histogram according to the present invention can more adaptively perform human tracking using different target color histograms according to the human poses, instead of applying only one target color histogram to the tracking process of one person, such that the accuracy of human tracking can be increased. The human tracking method includes performing color space conversion of input video data; calculating a state equation of a particle based on the color-space conversion data; calculating the state equation, and calculating human pose-adaptive observation likelihood; resampling the particle using the observation likelihood, and estimating a state value of the human; and updating a target color histogram.