Invensense, Inc.

Nvidia Corporation

Seiko Epson Corporation

Magnachip Semiconductor, Inc.

Mediatek Singapore Pte. Ltd.

Vivante Corporation

Ic Media Corporation

Shang-Hung Lin

Yuan Zheng

William Kerry Keal

David Sachs

Steve Nasiri

Ignatius B Tjandrasuwita

Joseph Jiang

Erik E Anderson

Kerry Keal

Anjia Gu

Xin Ru Wang

Brian Keith Cabral

Steven Saeed Nasiri

James Robert Lim

Takashi Nakayama

Vinayak Jayaram Pore

Mansur Kiadeh

Sohei Takemoto

Hu He

Shashank Garg

Sheena F Shi

Richard Li

Bin Fu

Alex Castro

Ashish Thakur

Ben S Wu

James Cape

Hung-Min Jen

Takashi Nakayami

A convolution neural network (CNN) model is trained and pruned at a pruning ratio. The model is then trained and pruned one or more times without constraining the model according to any previous pruning step. The pruning ratio may be increased at each iteration until a pruning target is reached. The model may then be trained again with pruned connections masked. The process of pruning, retraining, and adjusting the pruning ratio may also be repeated one or more times with a different pruning target.

Pruning and retraining method for a convolution neural network

An inertial measurement system is disclosed. The inertial measurement system has an accelerometer processing unit that generates a calibrated accelerometer data. The inertial measurement system further includes a magnetometer processing unit generates a calibrated magnetometer data, and a gyroscope processing unit generates a calibrated gyroscope data. Using the calibrated accelerometer data, the calibrated magnetometer data, and the calibrated gyroscope data, the inertial measurement system generates a heading angle error indicative of the accuracy of the heading angle error.

Heading confidence interval estimation

Systems and methods are disclosed for detecting when a magnetic anomaly may impact the quality of data being output by a magnetometer. A plurality of detection algorithms may be performed in parallel on the sensor data. Further, indication of a anomaly from one or a combination of the detection algorithms may cause the magnetometer data to have a reduced contribution in any sensor fusion operation or may be omitted from a sensor fusion operation as desired.

Systems and methods for detecting and handling a magnetic anomaly

Systems, methods, and apparatus for performing deduced reckoning navigation without a constraint relationship between orientation of a sensor platform and a direction of travel of an object are described herein. A sensor fusion component can be configured to receive data from sensors of a sensor platform coupled to a pedestrian; and generate world coordinate information based on the data. Further, a gait recognition component can be configured to record one or more walking patterns of the pedestrian in a training database; and determine whether the world coordinate information is associated with a walking pattern of the one or more walking patterns. Furthermore, a position estimation component can be configured to estimate a position of the pedestrian based on the world coordinate information if the world coordinate information is associated with the walking pattern, regardless of an orientation of the sensor platform with respect to the position of the pedestrian.

Deduced reckoning navigation without a constraint relationship between orientation of a sensor platform and a direction of travel of an object

Described herein is an intelligent remote controlling device (e.g. a mobile phone). The device can include a six-axis motion sensor to accurately track three dimensional hand motions. For example, the sensors can include a three-axis accelerometer and a three-axis gyroscope. The remote control device can also include a processing unit integrated with the motion sensors in a single module. The processing unit can convert data regarding the hand motion to data regarding a cursor motion for a cursor that will be displayed on a screen of an electronic device. The processing unit can be integrated with the motion sensors in a single module (e.g. an integrated circuit chip (IC)). The processing unit can include at least two modes of functionality corresponding to different types of hand motion: a one to one mode where the cursor directly tracks the hand motion and a non-linear mode that filters data from the motion sensors to eliminate hand jitter.

High fidelity remote controller device for digital living room

A method, system and computer readable medium for calibrating an accelerometer in a portable device is disclosed. The method, system and computer readable medium comprises receiving data from the accelerometer, and providing accelerometer samples from the data based upon one or more selection rules that adaptively selects data that satisfy certain criteria. The method system and computer readable medium also includes fitting the accelerometer samples to a mathematical mode. The method system and computer readable medium further includes providing a bias of the accelerometer based upon a center of the mathematical model.

In-use automatic calibration methodology for sensors in mobile devices

Systems and methods for human hand gesture recognition through a training mode and a recognition mode are disclosed. In the training mode, a user can move a handheld device with a hand gesture intended to represent a command. Sensors within the handheld device can record raw data, which can be processed to obtain a set of values corresponding to a set of discrete features, which is stored in a database and associated with the intended command. The process is repeated for various hand gestures representing different commands. In the recognition mode, the user can move the handheld device with a hand gesture. A computer system can compare a set of values corresponding to a set of discrete features derived from the hand gesture with the sets of values stored in the database, select a command with the closest match and displays and/or executes the command.

Handheld computer systems and techniques for character and command recognition related to human movements

Methods and systems for reducing or eliminating distortion in an image are described. The approach generally involves determining the distortion introduced by a lens, and modifying a captured image to reduce that distortion. In one embodiment, the distortion information associated with a lens is determined. The distortion information is stored. A captured image taken by that lens is processed, with reference to the distortion information.

Image distortion correction

A device disclosed herein can include a six-axis motion sensor to accurately track three dimensional hand motions. For example, the sensors can include a three-axis accelerometer and a three-axis gyroscope. The remote control device can also include a processing unit integrated with the motion sensors in a single module. The processing unit can convert data regarding the hand motion to data regarding a cursor motion for a cursor that will be displayed on a screen of an electronic device. The processing unit can be integrated with the motion sensors in a single module (e.g. an integrated circuit chip (IC)).

Described herein is an intelligent remote controlling device. The device can include a six-axis motion sensor to accurately track three dimensional hand motions. For example, the sensors can include a three-axis accelerometer and a three-axis gyroscope. The remote control device can also include a processing unit integrated with the motion sensors in a single module. The processing unit can convert data regarding the hand motion to data regarding a cursor motion for a cursor that will be displayed on a screen of an electronic device. The processing unit can be integrated with the motion sensors in a single module. The processing unit can include at least two modes of functionality corresponding to different types of hand motion: a one to one mode where the cursor directly tracks the hand motion and a non-linear mode that filters data from the motion sensors to eliminate hand jitter.

A flicker band automated detection system and method are presented. In one embodiment an incidental motion mitigation exposure setting method includes receiving image input information; performing a motion mitigating flicker band automatic detection process; and implementing exposure settings based upon results of the motion mitigating flicker band automatic detection process. The auto flicker band detection process includes performing a motion mitigating process on an illumination intensity indication. Content impacts on an the motion mitigated illumination intensity indication are minimized. The motion mitigated illumination intensity indication is binarized. A correlation of the motion mitigated illumination intensity and a reference illumination intensity frequency is established.

Flicker band automated detection system and method

An image processor in an image capture device compensates for the effects of undesirable camera shakes occurring during video capture The image processor receives a pair of source frames representing images of a scene, generates a pair of subsampled frames from the source frames, and computes a coarse displacement of the captured image due to camera shakes by comparing the two subsampled frames. The image processor may then refine the determined coarse displacement by comparing the two source frames and a bound determined by an extent of subsampling, and compensate for the displacement accordingly. Display aberrations such as blank spaces caused due to shifting are also avoided by displaying only a portion of the captured image and shifting the displayed portion to compensate for camera shake. The image processor also recognizes displacements due to intentional camera movement, and does not correct for such displacements.

Compensating for undesirable camera shakes during video capture

A technique for reducing artifacts in a digital image, in accordance with one embodiment, includes receiving a stream of raw filter pixel data representing the image. The raw filter pixel data is interpolating to produce red, green-on-red row, green-on-blue row and blue pixel data for each pixel. An artifact in one or more given pixels is reduced as a function of a difference between the green-on-red row and green-on-blue row pixel data of each of the given pixels to generate adjusted interpolated pixel data.

Growing community of inventors

San Jose, CA, United States of America

Shang-Hung Lin