Seiko Epson Corporation

Siemens Medical Solutions Usa, Inc.

Jing Xiao

Yuanyuan Ding

Derek Shiell

Chenyu Wu

Mithun Das Gupta

Kar-Han Tan

Anoop Kumar Bhattacharjya

Jinjun Wang

Ali Zandifar

Chunyu Gao

Che-Bin Liu

Seung-il Huh

Dorin Comaniciu

Bogdan Georgescu

Xiang Sean Zhou

Kenji Fukasawa

Guang Chen

Xianwang Wang

Zhaowen Wang

Sanjeev Kumar

Fan Jiang

A robot is made to recognize and manipulate different types of cable harnesses in an assembly line. This is achieved by using a stereo camera system to define a 3D cloud of a given cable harness. Pose information of specific parts of the cable harness are determined from the 3D point cloud, and the cable harness is then re-presented as a collection of primitive geometric shapes of known dimensions, whose positions and orientations follow the spatial position of the represented cable harness. The robot can then manipulate the cable harness by using the simplified representation as a reference.

2D/3D localization and pose estimation of harness cables using a configurable structure representation for robot operations

An indexed hierarchical tree search structure implements incremental tree modification to add new registrants to the tree without requiring reconstruction of the tree. Only data within a leaf node that receives new data is used in the incremental tree modification. Consequently, the original data set used in the creation of the hierarchical tree may be deleted after construction of the tree.

Constructing incremental tree model for vein image recognition and authentication

A stereo camera system produces a stereo image pair of a cable harness, which is used to define a 3D point cloud of the cable harness at its current pose position. Pose information of specific parts of the cable harness are determined from the 3D point cloud, and the cable harness is then re-presented as a collection of primitive geometric shapes of known dimensions, whose positions and orientations follow the spatial position and orientation of the imaged cable harness. The length, position and number of geometric shapes are atomically determined from a 2D image segmentation of one of the images in the stereo image pair.

Geodesic distance based primitive segmentation and fitting for 3D modeling of non-rigid objects from 2D images

Aspects of the present invention include systems and methods for generating a novel view interpolation. In embodiments, feature correspondences and geometrical contexts are used to find additional correspondences based on the assumption of the local linear transformation. The accuracy and the number of correspondence matches may be improved by iterative refinement. Having obtained a set of correspondences, a novel view image can be generated.

Context constrained novel view interpolation

A magnetic imaging systems produces magnetic magnitude images using magnetic sensors capable of determining only the absolute value of a detected magnetic field and provide no information regarding the positive or negative sign of the detected magnetic image. A 2D dipole location is determines the 2D location of a dipole within a magnetic magnitude image by finding the minimum of the derivative of the absolute value of the magnetic field. This 2D dipole location is then used to determine the 3D position and momentum of a current dipole responsible for the observed magnetic magnitude image. The current dipole is used to generate a magnetic image that incorporate positive and negative sign information.

2D dipole localization using absolute value of MCG measurements

A system identifies double-layer dipoles in a magnetic image by defining two-dimensional patches of distributed point charges that simulate the double-layer dipoles. The geometric center of a two-dimensional patch is used as the location of an equivalent dipole moment of the double-layer dipole. The momentum of the equivalent dipole moment is determined by submitting the magnetic image to a dipole construction system that identifies the location and momentum based on the submitted magnetic image.

Continuous modeling for dipole localization from 2D MCG images with unknown depth

A magnetocardiogram (MCG) system with reduced noise artifacts is produced by first creating high-resolution image representations of low-resolution measurements obtained with a magnetic field sensor unit. The high-resolution image representations are created by use of a PCA model that has been trained using a library of ideal, no-noise, high-resolution images. The Biot-Sarvart Law is then used to create a 3D model of a current impulse, given the high-resolution image representations. From the 3D current impulse model, ideal sensor unit measurements as they would have been obtained using a theoretical sensor unit observing the 3D current impulse model are synthesized.

Denoise MCG measurements

Ultrasound images are stitched together by extracting target image regions within each ultrasound image, identifying interest points for the extracted target images and then identifying interest point correspondences between the interest points of all the extracted target image regions. The interest point correspondence is used to define a common, word coordinate for all extracted target images, and the extracted target images are warped onto the world coordinates. A distance weight, w, gradient weight w, and histogram weight wis then determined for each pixel of the warped, target images. The three weights are then combined and normalized across all the warped, target images. The combined, normalized weights are used in the blending and stitching of the warped, target images.

Intelligent weighted blending for ultrasound image stitching

Aspects of the present invention comprise generating and using Multi-Order Contextual co-Occurrence (MOCO) descriptors to implicitly model the high level context using detection responses from a baseline object detector. In embodiments, a 1-order context feature is computed as a set of randomized binary comparisons on a response map of the baseline object detector. The statistics of the 1-order binary context features are further calculated to construct a higher-order co-occurrence descriptor, which, in embodiments, may be combined with other features such as the 0-order context features and/or the 1-order features to form the MOCO. In embodiments, combining the MOCO feature with the original image feature, the baseline object detector may be evolved to a stronger context aware detector.

Detector evolution with multi-order contextual co-occurrence

Linear candidate lines at different angles are used to determine an average intensity for each pixel level of an ultrasound image. The resultant average intensities are collected into a histogram, and the histogram is used to determine the depth positions of tissue boundary lines within an ultrasound image.

Orientation-aware average intensity histogram to indicate object boundary depth in ultrasound images

Fatty tissue boundary depths and muscle tissue boundary depths are identified in an ultrasound image by first creating an average intensity histogram of the ultrasound image. The histogram has a plurality of peaks, but has the characteristic that one of its peaks corresponds to a fat boundary depth, and a second of its peaks corresponds to a muscle boundary depth. A first classifier based solely on the local-characteristics of individual peaks is used to identify a first fat tissue depth. A second classifier trained to find a muscle depth given a fat depth, receives the output from the first classifier and identifies an output muscle tissue depth relative to the first fat tissue depth. A third classifier trained to find a fat boundary depth given a muscle boundary depth, receives the output muscle tissue depth and outputs a second fat boundary depth.

Ultrasound image object boundary localization by intensity histogram classification using relationships among boundaries

A specific item within an item class is identified by defining sets of descriptor data from a training library. The collected descriptor data is grouped and organized into a hierarchical tree, where each leaf node is defined by relations between corresponding parts of the descriptor data. Registrable sets of descriptor data are then identified from a collection of registrable samples. The registrable sets of descriptors are sorted into the hierarchical tree. When an input sample to be identified is received, a test set of descriptor data is generated from the input sample. The test set is then sorted into the hierarchical tree. Each leaf node that receives a part of the test set provides a vote for the registered samples it contains. The registered sample with the most votes is deemed a match for the input sample.

Fast and robust classification algorithm for vein recognition using infrared images

Aspects of the present invention include feature point matching systems and methods. In embodiments, a tree model is used to find candidate matching features for query feature points. In embodiments, the tree model may be pre-learned using a set of sample images, or alternatively, the tree model may be constructed using one or more of the input images. In embodiments, features in one of the stereo images are registered with the tree model, and then features from the other stereo image are queried through the tree model to identify their correspondences in the registered stereo image. As compared to prior brute force matching methodologies, embodiments disclosed herein reduce the complexity and calculation time for determining matching feature points in stereo images.

Tree-model-based stereo matching

Systems and methods for developing and using adaptive threshold values for different input images for object detection are disclosed. In embodiments, detector response histogram-based systems and methods train models for predicting optimal threshold values for different images. In embodiments, when training the model, an optimal threshold value for an image is defined as the value that maximizes the reduction of false positive image patches while preserving as many true positive image patches as possible. Once trained, the model may be used to set different threshold values for different images by inputting a detector response histogram for the image patches of an image into the model to determine a threshold value for detection.

Growing community of inventors

Cupertino, CA, United States of America

Jing Xiao