Abstract: A method of tracking a target includes receiving from a source an observed depth image of a scene including the target. Each pixel of the observed depth image is labeled as either a foreground pixel belonging to the target or a background pixel not belonging to the target. Each foreground pixel is labeled with body part information indicating a likelihood that that foreground pixel belongs to one or more body parts of the target. The target is modeled with a skeleton including a plurality of skeletal points, each skeletal point including a three dimensional position derived from body part information of one or more foreground pixels.

Abstract: A method of tracking a target includes receiving an observed depth image of the target from a source and obtaining a posed model of the target. The model is rasterized into a synthesized depth image, and the pose of the model is adjusted based, at least in part, on differences between the observed depth image and the synthesized depth image.

Abstract: Systems and methods for estimating a posture of a body part of a user are disclosed. In one disclosed embodiment, an image is received from a sensor, where the image includes at least a portion of an image of the user including the body part. The skeleton information of the user is estimated from the image, a region of the image corresponding to the body part is identified at least partially based on the skeleton information, and a shape descriptor is extracted for the region and the shape descriptor is classified based on training data to estimate the posture of the body part.

Abstract: A method of tracking a target includes receiving from a source a depth image of a scene including the human subject. The depth image includes a depth for each of a plurality of pixels. The method further includes identifying pixels of the depth image that belong to the human subject and deriving from the identified pixels of the depth image one or more machine readable data structures representing the human subject as a body model including a plurality of shapes.

Abstract: Embodiments are disclosed herein that relate to generating a decision tree through graphical processing unit (GPU) based machine learning. For example, one embodiment provides a method including, for each level of the decision tree: performing, at each GPU of the parallel processing pipeline, a feature test for a feature in a feature set on every example in an example set. The method further includes accumulating results of the feature tests in local memory blocks. The method further includes writing the accumulated results from each local memory block to global memory to generate a histogram of features for every node in the level, and for each node in the level, assigning a feature having a lowest entropy in accordance with the histograms to the node.

Abstract: A method of tracking a target includes receiving an observed depth image of the target from a source and obtaining a posed model of the target. The model is rasterized into a synthesized depth image, and the pose of the model is adjusted based, at least in part, on differences between the observed depth image and the synthesized depth image.

Abstract: A method of tracking a target includes receiving an observed depth image of the target from a source and analyzing the observed depth image with a prior-trained collection of known poses to find an exemplar pose that represents an observed pose of the target. The method further includes rasterizing a model of the target into a synthesized depth image having a rasterized pose and adjusting the rasterized pose of the model into a model-fitting pose based, at least in part, on differences between the observed depth image and the synthesized depth image. Either the exemplar pose or the model-fitting pose is then selected to represent the target.

Abstract: Systems and methods for storing high dynamic range image data in a low dynamic range format may be used to store the high dynamic range image data in less memory. The memory bandwidth needed to access the high dynamic range data is reduced and processing performance may be improved when performance is limited by memory bandwidth. The high dynamic range image data is scaled and compressed into a low dynamic range format for storage in a render target. If the compressed high dynamic range image data contains multiple data samples per pixel, the data may be processed to produce filtered compressed high dynamic range image data with only one sample per pixel. The high dynamic range image may be reconstructed from the low dynamic range format data and further processed as high dynamic range format data for a range of applications.

Abstract: A method of tracking a target includes receiving an observed depth image of the target from a source and analyzing the observed depth image with a prior-trained collection of known poses to find an exemplar pose that represents an observed pose of the target. The method further includes rasterizing a model of the target into a synthesized depth image having a rasterized pose and adjusting the rasterized pose of the model into a model-fitting pose based, at least in part, on differences between the observed depth image and the synthesized depth image. Either the exemplar pose or the model-fitting pose is then selected to represent the target.

Abstract: One embodiment of the present invention sets forth an improved method for computing a cube map blur function. The method begins with a rendered cube map of the surrounding scene using conventional environment rendering techniques. The method then proceeds with three successive cylindrical blurs around each axis of a coordinate frame. The three blur operations accumulate results from each predecessor operation for the different pixels of the cube map, thereby generating a high quality cube map blur. One advantage of this technique is that a relatively low computational effort yields a blur function involving a relatively large number of source pixels for each resulting pixel. Therefore, the resulting cube map can be computed in real-time and is suitable for use in a wide range of lighting effects.

Abstract: A target tracking method includes modeling the target in a first frame with a first frame iteration of a machine-readable model and receiving an observed depth image of a second frame of a scene including the target. The first frame iteration of the machine-readable model is then adjusted into a second frame iteration of the machine-readable model based on the observed depth image of the second frame.

Abstract: A method of tracking a target includes receiving from a source an observed depth image of a scene including the target. Each pixel of the observed depth image is labeled as either a foreground pixel belonging to the target or a background pixel not belonging to the target. Each foreground pixel is labeled with body part information indicating a likelihood that that foreground pixel belongs to one or more body parts of the target. The target is modeled with a skeleton including a plurality of skeletal points, each skeletal point including a three dimensional position derived from body part information of one or more foreground pixels.

Abstract: A method of tracking a target includes classifying a pixel having a pixel address with one or more pixel cases. The pixel is classified based on one or more observed or synthesized values. An example of an observed value for a pixel address includes an observed depth value obtained from a depth camera. Examples of synthesized values for a pixel address include a synthesized depth value calculated by rasterizing a model of the target; one or more body-part indices estimating a body part corresponding to that pixel address; and one or more player indices estimating a target corresponding to that pixel address. One or more force vectors are calculated for the pixel based on the pixel case, and the force vector is mapped to one or more force-receiving locations of the model representing the target to adjust the model representing the target into an adjusted pose.

Abstract: A method of tracking a target includes receiving an observed depth image of the target from a source and analyzing the observed depth image with a prior-trained collection of known poses to find an exemplar pose that represents an observed pose of the target. The method further includes rasterizing a model of the target into a synthesized depth image having a rasterized pose and adjusting the rasterized pose of the model into a model-fitting pose based, at least in part, on differences between the observed depth image and the synthesized depth image. Either the exemplar pose or the model-fitting pose is then selected to represent the target.

Abstract: A target tracking method includes representing a human target with a machine-readable model configured for adjustment into a plurality of different poses. The machine-readable model includes a plurality of joints, including one or more magnetism joints, and each joint has a three-dimensional world space position. The method further includes receiving an observed depth image of the human target from a source. The observed depth image includes a plurality of observed pixels. A magnetism body part is assigned to one or more of the plurality of observed pixels, and a magnetism joint position is estimated based on world space positions of the one or more observed pixels assigned the magnetism body part. A joint of the machine-readable model is then shifted toward the magnetism joint position.

Abstract: A target tracking method includes representing a human target with a machine-readable model including a plurality of skeletal points. The plurality of skeletal points are adjustable into a plurality of different legal poses. The method further includes receiving an observed depth image of the human target from a source and determining proposed positions for one or more of the skeletal points of the machine-readable model based on the observed depth image. The proposed positions of the skeletal points are then adjusted to comply with one or more rules if one or more of the proposed positions violates any of the one or more rules.

Abstract: A method of tracking a target includes receiving an observed depth image of the target from a source and obtaining a posed model of the target. The model is rasterized into a synthesized depth image, and the pose of the model is adjusted based, at least in part, on differences between the observed depth image and the synthesized depth image.

Abstract: A visual target tracking method includes representing a human target with a machine-readable model configured for adjustment into a plurality of different poses and receiving an observed depth image of the human target from a source. The observed depth image is compared to the model. A refine-z force vector is then applied to one or more force-receiving locations of the model to move a portion of the model towards a corresponding portion of the observed depth image if that portion of the model is Z-shifted from that corresponding portion of the observed depth image.

Abstract: A target tracking method includes representing a human target with a machine-readable model configured for adjustment into a plurality of different poses and receiving an observed depth image of the human target from a source. One or more push force vectors are applied to one or more force-receiving locations of the model to push the model in an XY plane towards a silhouette of the human target in the observed depth image when portions of the model are shifted away from the silhouette of the human target in the observed depth image. One or more pull force vectors are applied to one or more force-receiving locations of the model to pull the model in an XY plane towards the silhouette of the human target in the observed depth image when portions of the observed depth image are shifted away from the silhouette of the model.