Qeexo, Co.

Hewlett-packard Development Company, L.P.

Plantronics, Inc.

Robert Boschgmbh

Rajen Bhatt

Kui Zhang

Shitong Mao

Raviprakash Kandury

Michelle Tai

Geoffrey Newman

Christopher Harrison

Raghavendra Balavalikar Krishnamurthy

Yanfei Chen

Sang Won Lee

Stephen Paul Schaefer

James Stephen Miller

Jian David Wang

Matthew Baran

Naman Kumar

Yong Yan

Varun Ajay Kulkarni

David A Bryan

Hasan M Masoud Smeir

Karanpreet Singh

HasanMMasoud Smeir

Poojan Patel

Stephen Miller

Provided are various mechanisms and processes for automatic computer vision-based defect detection using a neural network. A system is configured for receiving historical datasets that include training images corresponding to one or more known defects. Each training image is converted into a corresponding matrix representation for training the neural network to adjust weighted parameters based on the known defects. Once sufficiently trained, a test image of an object that is not part of the historical dataset is obtained. Portions of the test image are extracted as input patches for input into the neural network as respective matrix representations. A probability score indicating the likelihood that the input patch includes a defect is automatically generated for each input patch using the weighted parameters. An overall defect score for the test image is then generated based on the probability scores to indicate the condition of the object.

Automated machine vision-based defect detection

A multi-camera video conference call system is provided with a plurality of cameras connected together over a communication network to generate a corresponding plurality of input frame images taken from different perspectives of a video conference room, where the multi-camera video conference call system detects one or more human heads for any meeting participants captured in the input frame images, generates a head bounding box which surrounds each detected human head, extracts a body bounding box which surrounds the detected human head and at least an upper body portion of a meeting participant belonging to the detected human head, generates a participant identification feature embedding from each body bounding box, and performs person re-identification processing on all generated participant identification feature embeddings to determine a count of the meeting participants in the video conference room.

Assignment of unique identifications to people in multi-camera field of view

A videoconference system acquires images during a videoconference. A still image may be analyzed to identify persons participating in the videoconference. A first frame is generated for each person and represents a subsection of an image that focuses on the face of that person. These first frames may include unpleasant components. Background segmentation is performed on each first frame to remove or diminish such unpleasant components. For example, a second frame for a person may be generated by identifying the background segments in the first frame and blurring or replacing such background segments with a virtual background. When the second frames are put together, the differences in the first frames are obscured making it more pleasing to view. A composite stream of the second frames is generated, per layout rules, and sent to the far side for improved viewing of the participants at the far side.

System and method for generating virtual background for people frames

A video conference call system is provided with a camera to generate an input frame image of a conference room, where the video conference call system detects a human head for each meeting participant captured in the input frame image by applying a machine learning human head detector model to said input image frame, generates a head bounding box which surrounds each detected human head and identifies a corresponding meeting participant, extracts a pixel width measure and pixel height measure from each head bounding box, and applies the extracted pixel width measure and pixel height measure to one or more reverse lookup tables to extract meeting room coordinates for each meeting participant identified by a corresponding head bounding box.

Estimation of human locations in two-dimensional coordinates using machine learning

A method may include obtaining, using a head detection model and for an image of a video stream, head detection information for heads detected in the image. The head detection information may include depth distances of the heads. Method may also include obtaining bounding boxes, where obtaining the bounding boxes may include obtaining head bounding boxes for the heads detected in the image, and combining at least two of the head bounding boxes into a combined bounding box according to the depth distances. Method may furthermore include creating, individually, head frame definitions for the bounding boxes, and processing the video stream using the head frame definitions.

Framing in a video system using depth information

Systems, computer-implemented methods, and computer program products that can facilitate calibrating a user activity model of a user device nodes are described. According to an embodiment, a method for calibrating a user activity model used by a mobile device can comprise receiving sensor data from a sensor of the mobile device. Further, applying a first weight to a first a first likelihood of a first occurrence of a first activity, wherein the first likelihood is determined by a first estimator of the user activity model by applying preconfigured criteria to the sensor data. The method can further comprise performing an action based on a determination of the first occurrence of the first activity, the determination being based on the first weight and the first likelihood of the first occurrence of the first activity.

Method and apparatus for calibrating a user activity model used by a mobile device

The number of persons in a still image acquired during a videoconference may change over time due to movement or persons entering or exiting a conference room. Minimizing the number of person identifiers that are tracked is beneficial for framing purposes. Typically, a stream of frames is sent to a far side, such as another endpoint device, for viewing. If the number of persons identified keeps changing, the composition of the stream of frames will be constantly changing as well. By using person identification and the use of timers, movement by the same person or a temporary change in a number or set of persons may be detected without making changes to the number of identifiers and persons being framed. It is only when a change has persisted past a period of time that changes are made to the frames, thereby improving the overall viewing experience.

Multi-person tracking and identification for robust framing experience

Disclosed are apparatus and methods for automatically training a sensor node to detect anomalies in an environment. At the sensor node, an indication is received to initiate training by the sensor node to detect anomalies in the environment based on sensor data generated by a sensor that resides on such sensor node and is operable to detect sensor signals from the environment. After training is initiated, the sensor node automatically trains a model that resides on the sensor to detect anomalies in the environment, and such training is based on the sensor data. After the model is trained, the model to detect anomalies in the environment is executed by the sensor node.

Method and system for training machine learning models on sensor nodes

A system for preventing private image data captured at an endpoint from being shared during a videoconference is provided. A user can select three-dimensional regions which will not be seen during a videoconference while areas in front of the designated regions remain viewable.

Securing image data from unintended disclosure at a videoconferencing endpoint

The disclosure facilitates fingerprint recognition, user authentication, and prevention of loss of control of personal information and identity theft. The disclosure also facilitates identifying spoofed fingerprint authentication attempts, and/or securing user touch sensitive devices against spoofed fingerprint authentication attempts.

Method and system to prevent identity theft for fingerprint recognition enabled touch screen devices

An object detection system for used in a vehicle includes an object detector is provided. One or more fisheye lens cameras coupled to an object detection system are positioned at various location of the vehicle for capturing a field of view (FOV) into an image. The image is split into multiple different set of perspective images. Each perspective images may include a portion of overlap having common or identical object of interest. The object detector classifies various objects of interest in the perspective images, estimates distance of objects of interest from the vehicle using camera calibration information stored in one of the camera, and transmits the sensed information that corresponds to the distance of interest to a processor. The processor includes sequence of instruction or object code either located in one of the object detection system, in-vehicle network, and an electronic device processes the sensed information and applies bounding boxes around objects of interest the vehicle or in the electronic device. One or more images incorporated the colored bounding boxes are transmitted to a display unit or screen for display as human readable format.

Growing community of inventors

McDonald, PA, United States of America

Rajen Bhatt