Intel Corporation

Mark Kliger

Shahar Fleishman

Gershom Kutliroff

Alon Lerner

Yaron Yanai

Maoz Madmony

Naomi Ken Korem

A system, article, and method of recurrent semantic segmentation for image processing by factoring historical semantic segmentation.

Method and system of recurrent semantic segmentation for image processing

Convolutional Neural Networks are applied to object meshes to allow three-dimensional objects to be analyzed. In one example, a method includes performing convolutions on a mesh, wherein the mesh represents a three-dimensional object of an image, the mesh having a plurality of vertices and a plurality of edges between the vertices, performing pooling on the convolutions of an edge of a mesh, and applying fully connected and loss layers to the pooled convolutions to provide metadata about the three-dimensional object.

Application of convolutional neural networks to object meshes

Techniques are provided for context-based 3D scene reconstruction employing fusion of multiple instances of an object within the scene. A methodology implementing the techniques according to an embodiment includes receiving 3D image frames of the scene, each frame associated with a pose of a depth camera, and creating a 3D reconstruction of the scene based on depth pixels that are projected and accumulated into a global coordinate system. The method may also include detecting objects, based on the 3D reconstruction, the camera pose and the image frames. The method may further include classifying the detected objects into one or more object classes; grouping two or more instances of objects in one of the object classes based on a measure of similarity of features between the object instances; and combining point clouds associated with each of the grouped object instances to generate a fused object.

Three dimensional scene reconstruction based on contextual analysis

Systems, apparatuses and/or methods may provide for generating a packing order of items within a container that consolidates the items into a reduced space. Items may be scanned with a three-dimensional (3D) imager, and models may be generated of the items based on the data from the 3D imager. The items may be located within minimal-volume enclosing bounding boxes, which may be analyzed to determine whether they may be merged together in one of their bounding boxes, or into a new bounding box that is spatially advantageous in terms of packing. If a combination of items is realizable and is determined to take up less space in a bounding box than the bounding boxes of the items considered separately, then they may be merged into a single bounding box. Thus, a spatially efficient packing sequence for a plurality of real objects may be generated to maximize packing efficiency.

Efficient packing of objects

Techniques are provided for 3D analysis of a scene including detection, segmentation and registration of objects within the scene. The analysis results may be used to implement augmented reality operations including removal and insertion of objects and the generation of blueprints. An example method may include receiving 3D image frames of the scene, each frame associated with a pose of a depth camera, and creating a 3D reconstruction of the scene based on depth pixels that are projected and accumulated into a global coordinate system. The method may also include detecting objects, and associated locations within the scene, based on the 3D reconstruction, the camera pose and the image frames. The method may further include segmenting the detected objects into points of the 3D reconstruction corresponding to contours of the object and registering the segmented objects to 3D models of the objects to determine their alignment.

3-dimensional scene analysis for augmented reality operations

Techniques related to pose estimation for an articulated body are discussed. Such techniques may include extracting, segmenting, classifying, and labeling blobs, generating initial kinematic parameters that provide spatial relationships of elements of a kinematic model representing an articulated body, and refining the kinematic parameters to provide a pose estimation for the articulated body.

Detection, tracking, and pose estimation of an articulated body

Techniques related to non-rigid transformations for articulated bodies are discussed. Such techniques may include repeatedly selecting target positions for matching a kinematic model of an articulated body, generating virtual end-effectors for the kinematic model and corresponding to the target positions, generating an inverse kinematics problem including a Jacobian matrix, and determining a change in kinematic model parameters based on the inverse kinematics problem until a convergence is attained.

Iterative closest point technique based on a solution of inverse kinematics problem

Techniques related to labeling component parts and detecting component properties in imaging data are discussed. Such techniques may include generating a feature vector including invariant features associated with an area of interest within an image of an object such as an image of a hand and providing a component label such as a hand part label for the area of interest based on an application of a machine learning classifier to the feature vector.

Growing community of inventors

Tel Aviv, Israel

Mark Kliger