126 results about "3D pose estimation" patented technology

The present invention is embodied in a video flashlight method. This method creates virtual images of a scene using a dynamically updated three-dimensional model of the scene and at least one video sequence of images. An estimate of the camera pose is generated by comparing a present image to the three-dimensional model. Next, relevant features of the model are selected based on the estimated pose. The relevant features are then virtually projected onto the estimated pose and matched to features of the image. Matching errors are measured between the relevant features of the virtual projection and the features of the image. The estimated pose is then updated to reduce these matching errors. The model is also refined with updated information from the image. Meanwhile, a viewpoint for a virtual image is selected. The virtual image is then created by projecting the dynamically updated three-dimensional model onto the selected virtual viewpoint.

A system, method, and computer program product for estimating upper body human pose are described. According to one aspect, a plurality of anatomical features are detected in a depth image of the human actor. The method detects a head, neck, and torso (H-N-T) template in the depth image, and detects the features in the depth image based on the H-N-T template. An estimated pose of a human model is estimated based on the detected features and kinematic constraints of the human model.

Techniques for performing accurate and automatic head pose estimation are disclosed. According to one aspect of the techniques, head pose estimation is integrated with a scale-invariant head tracking method along with facial features detected from a located head in images. Thus the head pose estimation works efficiently even when there are large translational movements resulting from the head motion. Various computation techniques are used to optimize the process of estimation so that the head pose estimation can be applied to control one or more objects in a virtual environment and virtual character gaze control.

Techniques are disclosed for the automatic recovery of two dimensional (2D) and three dimensional (3D) poses of multiple subjects interacting with one another, as depicted in a sequence of 2D images. As part of recovering 2D and 3D pose estimates, a pose recovery tool may account for constraints on positions of body parts of the first and second person resulting from the correlated activity. That is, individual subjects in the video are treated as mutual context for one another.

The present invention discloses a computer vision-based express parcel violent sorting identification method. The method comprises the following steps of: a depth camera-based pose estimation: a human pose estimation problem is converted into a problem of classifying depth image pixels captured by a depth camera, and human body pose estimation is realized by using a random forest method; human body three-dimensional pose relative spatial-temporal feature extraction: relative spatial-temporal positions of geometric elements of points, lines and surfaces formed by joints in three-dimensional poses and the measures of the change of the relative spatial-temporal positions are extracted and are adopted as the feature representations of the poses; and recurrent neural network-based violent sorting identification: modeling training is performed on the pose spatial-temporal relative features which are continuous in time and are extracted from the human body three-dimensional poses through a long and short memory model (LSTM), so that the identification of express parcel violent sorting behaviors can be realized.

Methods and systems for estimating a pose of a subject. The subject can be a human, an animal, a robot, or the like. A camera receives depth information associated with a subject, a pose estimation module to determine a pose or action of the subject from images, and an interaction module to output a response to the perceived pose or action. The pose estimation module separates portions of the image containing the subject into classified and unclassified portions. The portions can be segmented using k-means clustering. The classified portions can be known objects, such as a head and a torso, that are tracked across the images. The unclassified portions are swept across an x and y axis to identify local minimums and local maximums. The critical points are derived from the local minimums and local maximums. Potential joint sections are identified by connecting various critical points, and the joint sections having sufficient probability of corresponding to an object on the subject are selected.

Methods and apparatus are described for monocular 3D human pose estimation and tracking, which are able to recover poses of people in realistic street conditions captured using a monocular, potentially moving camera. Embodiments of the present invention provide a three-stage process involving estimating (10, 60, 110) a 3D pose of each of the multiple objects using an output of 2D tracking-by detection (50) and 2D viewpoint estimation (46). The present invention provides a sound Bayesian formulation to address the above problems. The present invention can provide articulated 3D tracking in realistic street conditions.

The present invention provides methods and apparatus for people detection and 2D pose estimation combined with a dynamic motion prior. The present invention provides not only 2D pose estimation for people in side views, it goes beyond this by estimating poses in 3D from multiple viewpoints. The estimation of poses is done in monocular images, and does not require stereo images. Also the present invention does not require detection of characteristic poses of people.

A method includes receiving, from an image capture device, a first image frame of a sequence of image frames. The method also includes estimating, at a processor, a camera pose corresponding to the first image frame by comparing the first image frame to a second image frame. The second image frame precedes the first image frame in the sequence of image frames. The method further includes estimating, at the processor, a refined camera pose corresponding to the first image frame by comparing the first image frame to a keyframe. The keyframe corresponds to a particular image frame that precedes the second image frame in the sequence of image frames.

The present invention provides a multi-camera-based collaborative visual SLAM method, in particular to a collaborative visual SLAM method using multiple cameras in a dynamic environment. The method of the invention allows the relative position and direction of the cameras to change with time, so that multiple cameras can move independently and be installed on different platforms, and solve problems related to camera pose estimation, map point classification and camera group management, etc. problem, so that the method works robustly in dynamic scenes and can reconstruct the 3D trajectories of moving objects. Compared with the existing SLAM method based on single camera, the method of the present invention is more accurate and robust, and can be applied to micro-robots and wearable augmented reality.

The present system discloses systems and methods for tracking planar shapes for augmented-reality (AR) applications. Systems for real-time recognition and camera six degrees of freedom pose-estimation from planar shapes are disclosed. Recognizable shapes can be augmented with 3D content. Recognizable shapes can be in form of a predefined library being updated online using a network. Shapes can be added to the library when the user points to a shape and asks the system to start recognizing it. The systems perform shape recognition by analyzing contour structures and generating projective invariant signatures. Image features are further extracted for pose estimation and tracking. Sample points are matched by evolving an active contour in real time.

The present system provides a tool to estimate the relative pose of a generic object with respect to a camera view-point by processing 2D images from a monocular camera in real-time. The capability of solving the pose estimation problem relies on the robust detection and matching in consecutive image frames of significant visual features belonging to an object of interest. To accomplish this, the system incorporates 3D modeling of the object of interest. In one embodiment, the shape of interest may be approximated by a parametric surface such as a cylinder, sphere, ellipsoid, or even complex non-parametric models. The system can restrain information retrieved at a 2D image level to estimate parameters about the pose. In operation, the accuracy of the 3D pose estimation of the object is a function of the degree of approximation of the selected model and the ability to select and track relevant features across consecutive image frames.

A camera pose estimation system is provided for estimating the position of a camera within an environment. The system may be configured to receive a 2D image captured by a camera within the environment, and interpret metadata of the 2D image to identify an estimated position of the camera. A synthetic 2D image from a 3D model of the environment may be rendered by a synthetic camera within the model at the estimated position. A correlation between the 2D image and synthetic 2D image may identify a 2D point of correlation, and the system may project a line from the synthetic camera through the 2D point on the synthetic 2D image rendered in an image plane of the synthetic camera such that the line intersects the 3D model at a corresponding 3D point therein. A refined position may be determined based on the 2D point and corresponding 3D point.

A method for estimating that posture of a three-dimensional human body based on a video stream. The method based on depth learning is used to estimate 3D pose of human body in video stream, which avoids many defects caused by two-dimensional vision analysis errors, and makes full use of the temporal relationship between video frames to improve the accuracy and real-time of 3D pose inference results of video stream. The method includes, for n(n >= 2) th frame of video, (1) inputting two-dimensional image of current frame and using shallow neural network module to generate image shallow map; 2)inputting a two-dimensional joint thermodynamic map of the human body generated by the (n-1)th frame and an image shallow map generated by the current frame to an LSTM module to generate a deep-levelcharacteristic map; 3) outputting the deep image characteristic map generated by the current frame to the residual module to generate a two-dimensional human body joint thermodynamic map of the current frame; 4) outputting the human body two-dimensional joint thermodynamic map of the current frame to a three-dimensional joint inference module for carrying out two-dimensional to three-dimensionalspatial mapping; and superimposing the three-dimensional human body joint thermodynamic map generated in each frame to generate a video stream of three-dimensional human body posture estimation.