Abstract: Provided are a visual positioning method, a non-transitory computer-readable storage medium and an electronic device. Surface normal vectors of a current image frame is obtained. A first transformation parameter between the current image frame and a reference image frame is determined, by projecting the surface normal vectors to a Manhattan coordinate system. A matching operation between feature points of the current image frame and feature points of the reference image frame is performed, and a second transformation parameter between the current image frame and the reference image frame is determined based on a matching result. A target transformation parameter is obtained, based on the first transformation parameter and the second transformation parameter. A visual localization result corresponding to the current image frame is output, based on the target transformation parameter.

Abstract: Apparatus for controlling a brightness of a display of a vehicle includes: an infrared illuminator, configured to emit infrared light in the vehicle; a near infrared (NIR) light sensing unit, configured to capture reflected infrared light; an image data processing unit, configured to analyze the reflected infrared light to generate a feedback; an imaging control unit, configured to adjust, in response to the feedback, one or more of a plurality of properties of the NIR light sensing unit, so that readouts of the NIR light sensing unit are within a first range, wherein the image data processing unit generates a calculated NIR intensity readout under the adjusted properties; a reconstruction unit, configured to reconstruct a human perceived brightness based on the calculated NIR intensity readout; and the display, configured to adjust the brightness. The NIR light sensing unit is a unit of a driver monitoring system (DMS).

Abstract: An apparatus for object detection around a vehicle includes an imaging device mounted on the vehicle and a computing device. The imaging device is used to capture a video of an exterior environment of the vehicle. The computing device is used to execute a DSFPN module in order to provide a DSFPN model detector, which functions as a perception unit to process and analyze the video captured by the imaging device and to estimate a scale, a location and categories of an object. The DSFPN model detector includes a bottom-up subnet provided with auxiliary prediction heads, and a top-down subnet provided with prediction heads. When the DSFPN model detector is performed in a model training stage, both the prediction heads and the auxiliary prediction heads are used. In a detection stage, only the prediction heads are used in the DSFPN model detector, and the auxiliary prediction heads are removed.

Abstract: Apparatus for vehicle driver recognition includes: a NIR LED illuminator, configured to emit NIR light in the vehicle; a NIR light sensing unit, configured to capture reflected NIR light; an image controlling and processing unit, configured to coordinate the NIR LED illuminator and the NIR light sensing unit, and analyze the reflected NIR light to generate an image; a face detector, configured to determine that a human face exists in the image, and identify a face region; a face feature extractor, configured to analyze the face region to extract a feature vector representing the face region; a face feature dictionary, configured to store existing feature vectors; a face retrieval system, configured to generate an identification result, indicating whether a similarity between the feature vector and any of the existing feature vectors is greater than a first threshold; and a user interface, configured to display the identification result.

Abstract: Apparatus for vehicle driver recognition includes: a NIR LED illuminator, configured to emit NIR light in the vehicle; a NIR light sensing unit, configured to capture reflected NIR light; an image controlling and processing unit, configured to coordinate the NIR LED illuminator and the NIR light sensing unit, and analyze the reflected NIR light to generate an image; a face detector, configured to determine that a human face exists in the image, and identify a face region; a face feature extractor, configured to analyze the face region to extract a feature vector representing the face region; a face feature dictionary, configured to store existing feature vectors; a face retrieval system, configured to generate an identification result, indicating whether a similarity between the feature vector and any of the existing feature vectors is greater than a first threshold; and a user interface, configured to display the identification result.

Abstract: Apparatus for controlling a brightness of a display of a vehicle includes: an infrared illuminator, configured to emit infrared light in the vehicle; a near infrared (NIR) light sensing unit, configured to capture reflected infrared light; an image data processing unit, configured to analyze the reflected infrared light to generate a feedback; an imaging control unit, configured to adjust, in response to the feedback, one or more of a plurality of properties of the NIR light sensing unit, so that readouts of the NIR light sensing unit are within a first range, wherein the image data processing unit generates a calculated NIR intensity readout under the adjusted properties; a reconstruction unit, configured to reconstruct a human perceived brightness based on the calculated NIR intensity readout; and the display, configured to adjust the brightness. The NIR light sensing unit is a unit of a driver monitoring system (DMS).

Abstract: Apparatus and method for measuring physiological information of a living subject in a vehicle are provided. The apparatus includes an illumination unit configured to emit light at a wavelength and illuminate a region of interest (ROI) of the living subject with the emitted light; a light sensing unit configured to remotely acquire signals of light in a range of wavelengths reflected from the ROI of the living subject responsive to the illumination; a remote photoplethysmography (PPG) extraction unit configured to extract PPG signals from the acquired signals of light; a PPG processing unit configured to process the extracted PPG signals in the form of vital signs and subsequently determine a physical and psychological status of the living subject according to the vital signs; and a control unit configured to activate the illumination unit in a time-varying pattern and to activate the light sensing unit accordingly.

Abstract: Methods, systems, and computer programs are presented for training a deep neural network (DNN). One method includes an operation for training a predecessor network defined for image recognition of items, where parameters of a predecessor classifier are initialized with random numbers sampled from a predetermined distribution, and the predecessor classifier utilizes an image-classification probability function without bias. The method further includes an operation for training a successor network defined for image recognition of items in a plurality of classes, where parameters of a successor classifier are initialized with parameters learned from the predecessor network, and the successor classifier utilizes the image-classification probability function without bias. Further, the method includes operations for receiving an image for recognition, and recognizing the image utilizing the successor classifier.

Abstract: Apparatus and method for measuring physiological information of a living subject in a vehicle are provided. The apparatus includes an illumination unit configured to emit light at a wavelength and illuminate a region of interest (ROI) of the living subject with the emitted light; a light sensing unit configured to remotely acquire signals of light in a range of wavelengths reflected from the ROI of the living subject responsive to the illumination; a remote photoplethysmography (PPG) extraction unit configured to extract PPG signals from the acquired signals of light; a PPG processing unit configured to process the extracted PPG signals in the form of vital signs and subsequently determine a physical and psychological status of the living subject according to the vital signs; and a control unit configured to activate the illumination unit in a time-varying pattern and to activate the light sensing unit accordingly.

Abstract: Methods, systems, and computer programs are presented for a classifier that recognizes images when at least one class includes just a few training samples. One method includes identifying a training set containing samples, each sample associated with a class from many classes. Further, the method divides the training set into a base and a novel set based on the number of samples in each class, trains a first classifier with the base set, and trains a second classifier using the training set. The second classifier is trained with promotion of the novel set and based on minimizing a loss function that comprises a first term and a second term, the first term associated with a first summation for the samples of the training set, the second term associated with a second summation for the samples of the novel set. Further, the method classifies an item with the trained second classifier.

Abstract: Systems and methods for visual recognition via light weight neural network are disclosed. A method includes accessing an input matrix. The method includes processing the input matrix through a plurality of convolution layers from a neural network architecture, each convolution layer including a convolution layer kernel, to generate a processed matrix, the convolution layer kernel being a first square, a side dimension of the first square being an integer greater than or equal to two. The method includes processing, at the processing hardware, the processed matrix through at least one squeeze layer, the at least one squeeze layer including a squeeze layer kernel, to generate an output matrix, the squeeze layer kernel being a second square with a side dimension of one, the at least one squeeze layer replacing at least one convolution layer from the neural network architecture. The method includes providing a representation of the output matrix.

Abstract: Strategies for improved neural network fine tuning. Parameters of the task-specific layer of a neural network are initialized using approximate solutions derived by a variant of a linear discriminant analysis algorithm. One method includes: inputting training data into a deep neural network having an output layer from which output is generated in a manner consistent with one or more classification tasks; evaluating a distribution of the data in a feature space between a hidden layer and the output layer; and initializing, non-randomly, the parameters of the output layer based on the evaluated distribution of the data in the feature space.

Abstract: Methods, systems, and computer programs are presented for training a deep neural network (DNN). One method includes an operation for training a predecessor network defined for image recognition of items, where parameters of a predecessor classifier are initialized with random numbers sampled from a predetermined distribution, and the predecessor classifier utilizes an image-classification probability function without bias. The method further includes an operation for training a successor network defined for image recognition of items in a plurality of classes, where parameters of a successor classifier are initialized with parameters learned from the predecessor network, and the successor classifier utilizes the image-classification probability function without bias. Further, the method includes operations for receiving an image for recognition, and recognizing the image utilizing the successor classifier.

Abstract: Methods, systems, and computer programs are presented for training a deep neural network (DNN). One method includes an operation for training a predecessor network defined for image recognition of items, where parameters of a predecessor classifier are initialized with random numbers sampled from a predetermined distribution, and the predecessor classifier utilizes an image-classification probability function without bias. The method further includes an operation for training a successor network defined for image recognition of items in a plurality of classes, where parameters of a successor classifier are initialized with parameters learned from the predecessor network, and the successor classifier utilizes the image-classification probability function without bias. Further, the method includes operations for receiving an image for recognition, and recognizing the image utilizing the successor classifier.

Abstract: A method of creating a hierarchical dictionary comprises, with a processor, extracting a number of symbols from a first image, constructing a number of refinement dictionary entries based on the symbols, the refinement dictionary entries forming a refinement dictionary, grouping a number of the refinement dictionary entries into clusters to form a number of refinement dictionary entry clusters, and constructing a number of direct dictionary entries for each of the refinement dictionary entry clusters, the direct dictionary entries forming a direct dictionary.

Abstract: Methods, systems, and computer programs are presented for training a deep neural network (DNN). One method includes an operation for training a predecessor network defined for image recognition of items, where parameters of a predecessor classifier are initialized with random numbers sampled from a predetermined distribution, and the predecessor classifier utilizes an image-classification probability function without bias. The method further includes an operation for training a successor network defined for image recognition of items in a plurality of classes, where parameters of a successor classifier are initialized with parameters learned from the predecessor network, and the successor classifier utilizes the image-classification probability function without bias. Further, the method includes operations for receiving an image for recognition, and recognizing the image utilizing the successor classifier.

Abstract: A system and method for text line detection are described Examples include detection of symbols in an image received from an image-capturing device. In examples, for each of at least some of the symbols, neighboring symbols within a local region a given distance from the symbol are analyzed in order to determine a direction for a line in the local region. In examples, based on the determined directions for the lines, text lines in the image are identified.

Abstract: Methods, systems, and computer programs are presented for a classifier that recognizes images when at least one class includes just a few training samples. One method includes identifying a training set containing samples, each sample associated with a class from many classes. Further, the method divides the training set into a base and a novel set based on the number of samples in each class, trains a first classifier with the base set, and trains a second classifier using the training set. The second classifier is trained with promotion of the novel set and based on minimizing a loss function that comprises a first term and a second term, the first term associated with a first summation for the samples of the training set, the second term associated with a second summation for the samples of the novel set. Further, the method classifies an item with the trained second classifier.

Abstract: A user query may be evaluated to provide a result set. In case the results do not reflect the user's intent, the device may provide recourse options for adjusting the query in a manner that yields more desirable results, e.g., a suggestion at the top of the result set for a different spelling, or recommendations at the end of the results set for additional query techniques that may yield more accurate results. However, such presentation of recourse options may clutter the user interface and/or go unnoticed by the user. Instead, an adjusted query may be identified with an interpreted probability of reflecting the intent of the query. An adjustment option describing the adjusted query may be inserted into the result set, between a higher-probability first result and a lower-probability second result. Selection of the adjustment option may cause the adjusted query to be evaluated on behalf of the user.

Abstract: A system and method for text line detection are described Examples include detection of symbols in an image received from an image-capturing device. In examples, for each of at least some of the symbols, neighboring symbols within a local region a given distance from the symbol are analyzed in order to determine a direction for a line in the local regions In examples, based on the determined directions for the lines, text lines in the image are identified.