Microsoft Technology Licensing, LLC

Amazon Technologies, Inc.

Google Inc.

Zongyi Liu

Haithem Albadawi

Hai Wei

Bruce Ferry

Christopher Nuesmeyer

Eric C Young

Jun Zhao

Devesh Mishra

Han Yee Mimi Fung

Timothy Jesse Tien

Sameer Vinod Shah

Kun Chen

Yongjun Wu

Sriram Sethuraman

Lei Li

Charles Benjamin Waggoner

Raymond W Smith

Ayush Goyal

Srinivas Rajagopalan

Chris Reidy

Joshua Peter Ebenezer

Qing Xie

Simon Paul Lacasse

Shunan Li

Andrew Becker

Systems and methods are described herein for detecting image impairments indicating low quality images within video content. An image impairment detection (IMD) model can be generated with a training data set including images that include one or more image impairments. Video content comprising a set of video frames may be obtained and a video frame provided to the image impairment detection model as input. Output may be received from the image impairment detection model indicating that the video frame includes an image impairment artifact indicative of a particular image impairment. One or more operations may be executed based at least in part on the output received.

Techniques for detecting low image quality

Techniques for content-adaptive video sampling for automated video quality monitoring are described. As one example, a computer-implemented method includes receiving a request to train a machine learning model on a training video file comprising at least one labeled defect, performing an encode on the training video file to generate one or more compression features for each compressed frame of the training video file, training the machine learning model to identify a proper subset of candidate defect frames of the training video file based at least in part on the one or more compression features for each compressed frame of the training video file and the at least one labeled defect, receiving an inference request for an input video file, performing an encode on the input video file to generate one or more compression features for each compressed frame of the input video file, generating, by the machine learning model, a proper subset of candidate defect frames of the input video file based at least in part on the one or more compression features for each compressed frame of the input video file, and determining a defect in the input video file based at least in part on the proper subset of candidate defect frames of the input video file.

Content-adaptive video sampling for cost-effective quality monitoring

Techniques are described for detecting desynchronization between an audio stream and a video stream.

Detecting audio-video desyncrhonization

A wearable heart rate monitoring device includes an optical sensor configured to translate test light reflected from a wearer of the wearable heart rate monitoring device into a machine-readable heart rate signal. The wearable heart rate monitoring device also includes an elevation sensor configured to translate an elevation of the wearable heart rate monitoring device into a machine-readable elevation signal. The wearable heart rate monitoring device further includes a heart rate reporting machine configured to output an estimated heart rate based on at least the machine-readable heart rate signal and the machine-readable elevation signal.

Heart rate correction for relative activity strain

Methods and devices for blood pressure monitoring may include receiving one or more sensor measurements from the at least one sensor. The methods and devices may further include determining at least one of a first blood pressure indication using a first regression representation based on the one or more sensor measurements, a second blood pressure indication using a second regression representation based on the one or more sensor measurements, or a third blood pressure indication using a third regression representation based on the one or more sensor measurements. The methods and devices may include performing a blood pressure selection procedure using the first blood pressure indication, the second blood pressure indication, and the third blood pressure indication to determine an estimated blood pressure indication based on one or more classification characteristics. The methods and devices may further include transmitting the estimated blood pressure indication to the output device.

Blood pressure determinations

Methods and devices for heart rate monitoring may include determining whether an adjustment triggering condition has been met. Moreover, the methods and devices may include, in accordance with a determination that the adjustment triggering condition has been met, adjusting the second sensor indication to obtain an adjusted second sensor indication based at least on the difference between the first sensor indication and the second sensor indication. The methods and devices may further include, in accordance with a determination that the adjustment triggering condition has not been met, transmitting the second sensor indication to the signaling filter.

Photoplethysmography (PPG) indication discontinuity detection and correction

Systems, methods, and computer-readable media are disclosed for systems and methods for latency detection for streaming video. Example methods may include receiving a request to determine a time to first frame for video content, determining a streaming service and a playback device associated with the request, and determining a time at which playback of the content is initiated. Example methods include capturing a playback sequence of the video content, classifying individual frames of the playback sequence, where each classification is associated with a probability value indicative of a likelihood the frame is a first frame, analyzing classifications of the individual frames, and determining, based at least in part on the analysis of the classifications of the individual frames, the time to first frame for the video content.

Latency detection for streaming video

Techniques for systems and methods that provide for utilizing a robotic system to type commands that correspond to voice commands using a keyboard of a device are described herein. In embodiments, an image of a device may be received from a camera. A keyboard region of the device may be determined based on a keyboard detection algorithm that uses the image. One or more characters in a portion of the image that corresponds to the keyboard region of the device may be detected based on a character detection algorithm. The one or more characters may be grouped into one or more groups based on the portion of the image. A character of a portion of character associated with a group may be edited based on an error detection algorithm.

Voice controlled keyboard typing using computer vision

Examples are disclosed herein that relate to entity tracking. One examples provides a computing device comprising a logic processor and a storage device holding instructions executable by the logic processor to receive image data of an environment including a person, process the image data using a face detection algorithm to produce a first face detection output at a first frequency, determine an identity of the person based on the first face detection output, and process the image data using another algorithm that uses less computational resources of the computing device than the face detection algorithm. The instructions are further executable to track the person within the environment based on the tracking output, and perform one or more of updating the other algorithm using a second face detection output, and updating the face detection algorithm using the tracking output.

Intelligent assistant

A computationally-efficient method for a smart assistant computer to track a human includes receiving data from one or more sensors configured to monitor a physical environment. The data is computer-analyzed to recognize presence of a human in the physical environment, and upon confirming an identity of the human, a first level of computational resources of the smart assistant computer is dedicated to track the human. Upon failing to confirm the identity of the human while a known user is present, a second level of computational resources of the smart assistant computer, greater than the first level, is dedicated to determine the identity of the human. Upon failing to confirm the identity of the human while the known user is absent, a third level of computational resources of the smart assistant computer, is dedicated to determine the identity of the human.

Computationally-efficient human-identifying smart assistant computer

Methods and devices for correcting electrocardiogram (EKG) indication saturation may include receiving an EKG indication from at least one sensor and determining that a saturation condition has been met in response to receiving the EKG indication. Additionally, the methods and devices may include incrementing an EKG saturation level based on a determination that the saturation condition has been met. Moreover, the methods and devices may include determining whether the EKG saturation level satisfies an EKG saturation threshold. The methods and device may include disregarding at least a second EKG indication in accordance with a determination that the saturation level satisfies the EKG saturation threshold. The methods and devices may further include transmitting the second EKG indication to an output device in accordance with a determination that the saturation level does not satisfy the EKG saturation threshold.

Electrocardiogram (EKG) indication saturation correction

An optical heart rate sensor stores data indicating the timing of heart beats in a first-in-first-out rolling buffer. During a first condition, new data is added to the rolling buffer and adjusted such that a duration preceding a newly detected heart beat is closer to an average duration between each pair of consecutive heart beats stored in the rolling buffer if the duration preceding the newly detected heart beat differs from the average duration by more than a duration threshold. The rolling buffer is cleared when a number of data adjustments over a predetermined time period exceeds a counting threshold. The rolling buffer is then repopulated such that the duration between each pair of consecutive heart beats is equal to a most recent duration between consecutive heart beats that did not differ from the average duration by more than the duration threshold.

Data recovery for optical heart rate sensors

A wearable heart rate monitoring device includes an optical sensor configured to translate test light reflected from a wearer of the wearable heart rate monitoring device into a machine-readable heart rate signal. The wearable heart rate monitoring device also includes a motion sensor configured to translate motion of the wearable heart rate monitoring device into a machine-readable motion signal. The wearable heart rate monitoring device also includes a heart rate reporting machine, configured to determine a type of activity currently being performed by the wearer of the wearable heart rate monitoring device based at least in part on the machine-readable motion signal, and output an estimated heart rate based on at least the machine-readable heart rate signal and the type of activity.

Heart rate correction

A method of optical heart rate sensing includes receiving a motion frequency from a motion sensor and receiving an optical signal from an optical sensor. The motion frequency is then filtered from the optical signal, and an estimated heart rate frequency is determined based on the filtered optical signal. A heart rate is reported that is based on the motion frequency, but not based on the estimated heart rate frequency, when a local neighbor magnitude ratio of the estimated heart rate frequency is below a confidence threshold.

Growing community of inventors

Issaquah, WA, United States of America

Zongyi Liu