University of Illinois

Personify, Inc.

International Business Machines Corporation

Sony Corporation

Pesonify, Inc.

Minh Do

Sanjay J Patel

Quang Nguyen

Dennis Lin

Daniel P Dabbelt

Long Dang

Sumant Kowshik

Long Truong

Ali Jezu Tabatabai

Rogerio Schmidt Feris

Jinjun Xiong

Simion Venshtain

Cong Nguyen

Dhiraj Joshi

Raymond Alexander Yeh

Mert Dikmen

Greg Meyer

Cheung Ayeung

Ha T Nguyen

Khoi-Nguyen C Mac

Quang Il Nguyen

Benjamin Chidester

Quang H Nguyen

Daniel P Dabbeit

Techniques regarding action detection based on motion in receptive fields of a neural network model are provided. For example, one or more embodiments described herein can comprise a system, which can comprise a memory that can store computer executable components. The system can also comprise a processor, operably coupled to the memory, and that can execute the computer executable components stored in the memory. The computer executable components can comprise a motion component that can extract a motion vector from a plurality of adaptive receptive fields in a deformable convolution layer of a neural network model. The computer executable components can also comprise an action detection component that can generate a spatio-temporal feature by concatenating the motion vector with a spatial feature extracted from the deformable convolution layer.

Action detection by exploiting motion in receptive fields

A system for background image subtraction includes a computing device coupled with a 3D video camera, a processor of the device programmed to receive a video feed from the camera containing images of one or more subject that include depth information. The processor, for an image: segments pixels and corresponding depth information to identify foreground regions associated with a target user.

System for background subtraction with 3D camera

Embodiments disclose systems and methods for transmitting user-extracted video and content more efficiently by recognizing that user-extracted video provides the potential to treat parts of a single frame of a user-extracted video differently. An alpha mask of the image part of the user-extracted video is used when encoding the image part so that it retains a higher quality upon transmission than the remainder of the user-extracted video.

Transmitting video and sharing content via a network

Disclosed herein are methods and systems for classifying pixels as foreground using both short-range depth data and long-range depth data. One embodiment takes the form of a process that includes obtaining video data depicting at least a portion of a user. The process also includes obtaining short-range depth data associated with the video data. The process also includes obtaining long-range depth data associated with the video data. The video data, short-range depth data, and long-range depth data may be obtained via a single 3-D video camera. The process also includes classifying pixels of the video data as foreground based at least in part on both the short-range depth data and the long-range depth data. In some embodiments, classifying pixels of the video data as foreground comprises employing an alpha mask. The alpha mask may comprise binary foreground (hard) indicators. The alpha mask may comprise foreground-likelihood (soft) indicators.

Methods and systems for classifying pixels as foreground using both short-range depth data and long-range depth data

A system for background image subtraction includes a computing device coupled with a 3D video camera, a processor o&#xa3; the device programmed to receive a video feed from the camera containing images of one or more subject that include depth information. The processor, for an image: segments pixels and corresponding depth information into three different regions including foreground (FG), background (BG), and unclear (UC); categorizes UC pixels as FG or BG using a function that considers the color and background history (BGH) information associated with the UC pixels and the color and BGH information associated with pixels near the UC pixels; examines the pixels marked as FG and applies temporal and spatial filters to smooth boundaries of the FG regions; constructs a new image by overlaying the FG regions on top of a new background; displays a video feed of the new image in a display device; and continually maintains the BGH.

A system is disclosed for executing depth image-based rendering of a 3D image by a computer having a processor and that is coupled with one or more color cameras and at least one depth camera. The color cameras and the depth camera are positionable at different arbitrary locations relative to a scene to be rendered. In some examples, the depth camera is a low resolution camera and the color cameras are high resolution. The processor is programmed to propagate depth information from the depth camera to an image plane of each color camera to produce a propagated depth image at each respective color camera, to enhance the propagated depth image at each color camera with the color and propagated depth information thereof to produce corresponding enhanced depth images, and to render a complete, viewable image from one or more enhanced depth images from the color cameras. The processor may be a graphics processing unit.

System for executing 3D propagation for depth image-based rendering

A color image and a depth image of a live video are received. Each of the color image and the depth image are processed to identify the foreground and the background of the live video. The background of the live video is removed in order to create a foreground video that comprises the foreground of the live video. A control input may be received to control the embedding of the foreground video into a second background from a background feed. The background feed may also comprise virtual objects such that the foreground video may interact with the virtual objects.

Systems and methods for embedding a foreground video into a background feed based on a control input

A system for background image subtraction includes a computing device coupled with a 3D video camera, a processor of the device programmed to receive a video feed from the camera containing images of one or more subject that include depth information. The processor, for an image: segments pixels and corresponding depth information into three different regions including foreground (FG), background (BG), and unclear (UC); categorizes UC pixels as FG or BG using a function that considers the color and background history (BGH) information associated with the UC pixels and the color and BGH information associated with pixels near the UC pixels; examines the pixels marked as FG and applies temporal and spatial filters to smooth boundaries of the FG regions; constructs a new image by overlaying the FG regions on top of a new background; displays a video feed of the new image in a display device; and continually maintains the BGH.

Transmitting video and sharing content via a network using multiple encoding techniques

A method, apparatus, system, and computer program product for of digital imaging. Multiple cameras comprising lenses and digital images sensors are used to capture multiple images of the same subject, and process the multiple images using difference information (e.g., an image disparity map, an image depth map, etc.). The processing commences by receiving a plurality of image pixels from at least one first image sensor, wherein the first image sensor captures a first image of a first color, receives a stereo image of the first color, and also receives other images of other colors. Having the stereo imagery, then constructing a disparity map and an associated depth map by searching for pixel correspondences between the first image and the stereo image. Using the constructed disparity map, captured images are converted into converted images, which are then combined with the first image, resulting in a fused multi-channel color image.

System and method for generating a depth map and fusing images from a camera array

A color image and a depth image of a live video are received. A user is extracted from the information of the color image and the depth image. Spurious depth vales may be corrected. Points or pixels of an image as seen from a viewpoint of a reference camera at a reference camera location are mapped to points of the image as would be seen from a viewpoint of a virtual camera at a virtual camera location. As such, a transformed color image is generated. Disoccluded pixels may be processed to address any gaps within the transformed color image.

Systems and methods for generating a virtual camera viewpoint for an image

An RGB color image and an infrared intensity image of a live video are received. The RGB color image is converted to a colorspace image comprising a channel corresponding to a brightness value. Each pixel of the converted colorspace image is evaluated to determine whether the brightness channel of the pixel exceeds a threshold value. If the brightness channel of the pixel exceeds the threshold value, the infrared intensity value of a corresponding pixel from the infrared intensity image is mixed into the pixel's channel value that corresponds to brightness. The converted colorspace image is converted back to an RGB color image.

Systems and methods for illumination correction of an image

A color image and a depth image of a live video are received. Each of the color image and the depth image are processed to identify a foreground, background, and an unknown region band of the live video. The unknown region band may comprise pixels between the foreground and the background. Further processing is performed to segment the pixels of the unknown region band between the foreground and the background. As such, processing is performed on the unknown region band in order to provide an improved user foreground video.

Systems and methods for accurate user foreground video extraction

An adaptive loop filter utilizing separable filters within a quadtree-based adaptive loop filter (QALF). The filters for at least the P and B pictures are replaced with a separable filter to provide Separable QALF (SQALF). In a preferred embodiment the filter comprises a Wiener filter as a best fit approximation of the non-separable filter. In response to decoding with the SQALF approach, computational complexity in the decoder is reduced without lowering perceived visual quality. Estimation of the separable filter is obtained in the encoder by iteratively minimizing the sum of squares error with the separable filter. The SQALF filtering is described within a video decoder.

Growing community of inventors

Urbana, IL, United States of America

Minh Do