A system and method for estimating a sequence of human poses in an unconstrained video. In the present invention, a unified two stage, tree-based, optimization problem is solved for which an efficient and exact solution exists. While the proposed method finds an exact solution, it does not sacrifice the ability to model the spatial and temporal constraints between body parts in the video frames on the unconstrained video.

Solutions for object tracking problems are presented by gathering images using one or more cameras, processing the gathered images to generate a directed acyclic graph, using the directed acyclic graph to determine a path cover that achieves maximum weight and satisfies one or more positive or negative constraints, and using the path cover to solve the object tracking problem. A first set of solutions utilizes trellis graphs, a second set of solutions employs a greedy approach, and a third set of solutions uses search algorithms.

A numerical parameter indicative of the degree of match between two retina images is produced by comparing two graphs obtained from the respective images. Each graph is composed of edges and vertices. Each vertex is associated with a location the corresponding retina image, and with descriptor data describing a part of the corresponding retina image proximate the corresponding location.

A method, computer readable storage medium, and system are disclosed for improving communication productivity in a conference between two or more subjects, wherein at least one of the two or more subjects participates in the conference from a first location and one or more of the two or more subjects participate in the meeting from a second location. The method includes capturing, at least one first three-dimensional (3D) stream of data and at least one second three-dimensional (3D) stream of data on each of the two or more subjects participating in the conference; generating a synchrony score for the two or more subjects, wherein the synchrony score is calculated by comparing time series of skeletal data of each of the two or more subjects to one another for a defined period of time; and using the synchrony score to generate an engagement index between the two or more subjects.

Features are disclosed for classifying pixels included in a digital image. Distance information from a pixel to structural reference points, such as skeletal joints, is generated. The distance information is then applied to a pixel classifier to identify one or more classifications for the pixel.

Some embodiments are directed to a method of cloud point simplification that includes implementing recursive spatial partitioning of the set of points into a hierarchy of clusters and, for each cluster, calculating a tangent plane estimate for the points in the cluster and deriving a confidence factor for the calculated tangent plane estimate. The method further includes identifying representative points within each cluster in the hierarchy and, for each representative point, defining a point-pair that consists of the representative point and a representative point of an immediate parent cluster; calculating a contraction error metric for each point-pair that is weighted by the inverse of the confidence factors calculated for the clusters associated with each representative point; and iteratively contracting the point-pair with the lowest contraction error metric, updating remaining point-pairs as a result of the contraction, and revising the contraction error metric of the updated point-pairs accordingly.

Methods, systems, and computer program products for parsing objects are provided herein. A method includes producing a plurality of versions of an image of an object derived from an input, wherein each version comprises one of one multiple resolutions of said image of said object; computing an appearance probability at each of a plurality of regions on the one or more lowest resolution versions of said plurality of versions of said image for at least one attribute for said object; determining a configuration of the at least one attribute in the one or more lowest resolution versions based on at least the appearance probability in each of the plurality of regions; and outputting said configuration.

In an example embodiment, a method, apparatus and computer program product are provided. The method includes facilitating receipt of an image of a scene and determining a graph based on connecting nodes of the image. The nodes are either pixels or superpixels of the image. The graph is determined by determining one or more connections of a node to one or more nodes belonging to a pre-defined image region around the node in the image. The connections are associated with edge weights that are determined based on at least one of similarity parameters and spatial distances between the node and the one or more nodes. The method includes determining disparity values at the nodes of the image based at least on performing tree based aggregation of a cost volume on the graph, where the cost volume is associated with the image and at least one view image of the scene.

A computer-implemented method and an apparatus for improving gesture recognition are described. The method comprises providing a reference model defined by a joint structure, receiving at least one image of a user, and mapping the reference model to the at least one image of the user, thereby connecting the user to the reference model for recognition of a set of gestures predefined for the reference model, when the gestures are performed by the user.

System and methods for relational time-series learning are provided. Unlike traditional time series forecasting techniques, which assume either complete time series independence or complete dependence, the disclosed system and method allow time series forecasting that can be performed on multivariate time series represented as vertices in graphs with arbitrary structures and predicting a future classification for data items represented by one of nodes in the graph. The system and methods also utilize non-relational, relational, temporal data for classification, and allow using fast and parallel classification techniques with linear speedups. The system and methods are well-suited for processing data in a streaming or online setting and naturally handle training data with skewed or unbalanced class labels.