Xerox Corporation

Conduent Business Services, LLC

The Procter Gamble Company

Palo Alto Research Center Inc.

Beilei Xu

Robert P Loce

Lalit Keshav Mestha

Wencheng Wu

Peter Paul

Shen-Ge Wang

Chu-Heng Liu

Edgar A Bernal

Survi Kyal

Stuart Alan Schweid

Graham Stephen Pennington

Raja Bala

Orhan Bulan

Joseph C Sheflin

Michael Branciforte

Jess Robert Gentner

Hao Wu

Yao Rong Wang

Yeqing Zhang

Himanshu Madhu

Son H Nguyen

Howard A Mizes

Thomas F Wade

David Jordan Lieberman

Safwan R Wshah

Reiner Eschbach

David Allen Mantell

Edul N Dalal

Florent C Perronnin

John C Handley

Jack T LeStrange

Ying-wei Lin

Jack Thomas Lestrange

Abu Saeed Islam

Jayant Kumar

Robert J Kleckner

Anthony Ang

Rakesh Suresh Kulkarni

Sandra Skaff

Waqas Sultani

Zhigang Fan

Palghat Srinivas Ramesh

Patrick Yasuo Maeda

Matthew Adam Shreve

Eric Scott Hamby

William J Nowak

R Enrique Viturro

Hui Ting Cheng

Robert Paul Herloski

Daniel W Costanza

Jon S McElvain

Michael D Thompson

Charles Michael Hains

Saurabh Prabhat

Paul Ronald Austin

Sohail A Dianat

Elliott A Eklund

Qun Li

Mary Ann Sprague

George Cunha Cardoso

Nancy Beth Goodman

Paul Jonathan Matts

Michael P Kehoe

Jeremy Lee Reitz

Timothy James Clark

Edward W Smith, Jr

Ankur Purwar

Kenneth R Miller

Craig John Saunders

Mark A Gwaltney

Patricia Swenton-Wall

Connie F Purdum

Yusuf Oguzhan Artan

Gill Rafael Tsouri

Rinku Gajera

Lee Anne Williams

Mark Cantelli

Megan Clar

Eric Harte

Xiaoxue Cheng

Faezeh Tafazzoli

Yi Liu

Jason Repko

According to exemplary methods of training a convolutional neural network, input images are received into a computerized device having an image processor. The image processor evaluates the input images using first convolutional layers. The number of first convolutional layers is based on a first size for the input images. Each layer of the first convolutional layers receives layer input signals comprising features of the input images and generates layer output signals that include signals from the input images and ones of the layer output signals from previous layers within the first convolutional layers. Responsive to an input image being a second size larger than the first size, additional convolutional layers are added to the convolutional neural network. The number of additional convolutional layers is based on the second size in relation to the first size. The additional convolutional layers are initialized using weights from the first convolutional layers. Feature maps comprising the layer output signals are created.

System and method for expanding and training convolutional neural networks for large size input images

Systems and methods for determining an apparent skin age of a person and providing customized skin care product recommendations. The system utilizes a computing device to mask facial macro features in an image of a person and then analyze the image with a convolutional neural network to determine an apparent skin age of the person. Determining the apparent skin age may include identifying at least one pixel that is indicative of skin age and utilizing the at least one pixel to provide the apparent skin age. The system may be used to determine a target skin age of a person, determine a skin care product or skin care regimen for achieving the target skin age, and provide an option for a user to purchase the product.

Systems and methods for determining apparent skin age

A computer-vision based method for validating an activity workflow of a human performer includes identifying a target activity. The method includes determining an expected sequence of actions associated with the target activity. The method includes receiving a video stream from an image capture device monitoring an activity performed by an associated human performer. The method includes determining an external cue in the video stream. The method includes associating a frame capturing the external cue as a first frame in a key frame sequence. The method includes determining an action being performed by the associated human performer in the key frame sequence. In response to determining the action in the key frame sequence matching an expected action in the target activity, the method includes verifying the action as being performed in the monitored activity. In response to not determining the action in the key frame sequence, the method includes generating an alert indicating an error in the monitored activity.

Computer-vision based process recognition of activity workflow of human performer

A method and system for domain adaptation based on multi-layer fusion in a convolutional neural network architecture for feature extraction and a two-step training and fine-tuning scheme. The architecture concatenates features extracted at different depths of the network to form a fully connected layer before the classification step. First, the network is trained with a large set of images from a source domain as a feature extractor. Second, for each new domain (including the source domain), the classification step is fine-tuned with images collected from the corresponding site. The features from different depths are concatenated with and fine-tuned with weights adjusted for a specific task. The architecture is used for classifying high occupancy vehicle images.

Multi-layer fusion in a convolutional neural network for image classification

A method for gait analysis of a subject performed periodically over time to detect changes in one or more gait characteristics. The method includes detecting and identifying a subject and analyzing the gait of the subject on a plurality of occasions. Analyzing the gait of the subject includes, a detecting body parts, generating a joint model depicting the location of the at least one joint in each of the at least two frames, using the joint model to segment a gait cycle for the at least one joint, and comparing the gait cycle to a threshold value to detect abnormal gait.

Computer vision system for ambient long-term gait assessment

A computer-implemented system and method for gait analysis of a subject. The method includes obtaining visual data from an image capture device positioned in front of or behind the subject, the plurality of image frames capturing at least one gait cycle of the gait of the subject, detecting one or more prescribed features within the plurality of image frames, analyzing each of the plurality of image frames to detect cross-frame stability of the one or more prescribed features, and segmenting the gait of the subject into at least one gait cycle based at least in part on the detected cross-frame stability of the one or more prescribed features.

System and method for automatic gait cycle segmentation

Methods, systems, and processor-readable media for detecting the side window of a vehicle. A spatial probability map can be calculated, which includes data indicative of likely side window locations of a vehicle in an image. A side window detector can be run with respect to the image of the vehicle to determine detection scores. The detection scores can be weighted based on the spatial probability map. A detected region of interest can be extracted from the image as extracted image patch. An image classification can then be performed with respect to the extracted patch to provide a classification that indicates whether or not a passenger is in the vehicle or no-passenger is in the vehicle.

Side window detection through use of spatial probability maps

Methods, systems, and processor-readable media for training data augmentation. A source domain and a target domain are provided, and thereafter an operation is performed to augment data in the source domain with transformations utilizing characteristics learned from the target domain. The augmented data is then used to improve image classification accuracy in a new domain.

Target domain characterization for data augmentation

The present disclosure provides a method of producing a personalized medical record, comprising: sensing capabilities of a receiving device; retrieving stock information; retrieving personalized information; combining at least a portion of the stock information and at least a portion of the personalized information into the personalized record; formatting the personalized record based on a combination of the capabilities of the receiving device and a user's preference; and, transmitting the formatted personalized record to the device.

Personalized medical record

A method, system, and apparatus for vehicle occupancy detection comprises collecting image data of a vehicle, creating a first value according to a plurality of characteristics of a driver position of the vehicle, and then creating at least one other value according to a plurality of characteristics of at least one other candidate occupant position in the vehicle. The characteristics of the driver position of the vehicle and candidate occupant position of the vehicle are compared in order to determine the vehicle occupancy.

Vehicle occupancy detection using passenger to driver feature distance

A system and method for detecting electronic device use by a driver of a vehicle including acquiring an image including a vehicle from an associated image capture device positioned to view oncoming traffic, locating a windshield region of the vehicle in the captured image, processing pixels of the windshield region of the image for computing a feature vector describing the windshield region of the vehicle, applying the feature vector to a classifier for classifying the image into respective classes including at least classes for candidate electronic device use and candidate electronic device non-use, and outputting the classification.

Machine learning approach for detecting mobile phone usage by a driver

What is disclosed is a system and method for determining respiration rate from a video of a subject. In one embodiment, a video is received comprising plurality of time-sequential image frames of a region of a subject's body. Features of pixels are extracted from that region from each image frame and vectors formed from these features. Each image frame has an associated feature vector. A N&#xd7;M video matrix of the vectors of length N is constructed such that a total number of columns M in the video matrix correspond to a time duration over which the subject's respiration rate is to be determined. The video matrix is processed to obtain a matrix of eigenvectors where principal axes of variations due to motion associated with respiration are contained in a first few eigenvectors. One eigenvector is selected from the first few eigenvectors. A respiration rate is obtained from the selected eigenvector.

System and method for determining respiration rate from a video

A system and method for assessing patient movement for Parkinson's disease includes capturing a video of a subject performing a finger tapping sequence comprising a predetermined number of open and close periods. According to an exemplary embodiment, a system and method includes extracting a region of interest for each frame of the video and generating a projection of the region of interest for each frame of the video using perpendicular vector projections in a direction or plurality of directions.

Automatic visual remote assessment of movement symptoms in people with parkinson's disease for MDS-UPDRS finger tapping task

What is disclosed is a system and method for extracting photoplethysmographic (PPG) signal (i.e., a cardiac signal) on a continuous basis from a time-series signals obtained from video images captured of a subject being monitored for cardiac function in a non-contact remote sensing environment involves the following. First, a time-series signal obtained from video images captured of a region of exposed skin where a photoplethysmographic (PPG) signal of a subject of interest can be registered. A sliding window is then used to define consecutive sequential segments of the time-series signal for processing. Each of the consecutive time-series signal segments is detrended such that low frequency variations and non-stationary components are removed. The detrended signals are processed to obtain, for each segment, a PPG signal. The PPG signal segments are then stitched together using a stitching method, as disclosed herein, to obtain a continuous PPG signal for the subject.

Continuous cardiac signal generation from a video of a subject being monitored for cardiac function

Growing community of inventors

Penfield, NY, United States of America

Beilei Xu