14 results about "Pose tracking" patented technology

This invention belongs to the field of on-orbit non-cooperative spacecraft pose tracking technology, specifically involving an online adaptive method to compensate for domain offset in on-orbit non-cooperative spacecraft pose tracking. Step 1: Represent the target spacecraft as 11 three-dimensional key points; extract the key point bounding boxes after projection, and crop to obtain the input RoI; train an asymmetric encoder-decoder network to regress the key point heatmap; Step 2: Design a dynamic memory library M based on a first-in-first-out queue; Step 3: The student model receives complete samples from M, and the teacher model receives masked samples from M; consistency constraints are implemented by minimizing the weighted MSE loss between the student and teacher predicted heatmaps; Step 4: Calculate the average value of the [CLS] token features of all training images as the source domain global class prototype, and align the source domain prototype with the newly calculated class prototypes of the teacher and student models. This invention makes it possible to achieve high-precision, highly robust, and highly practical real-time online tracking of non-cooperative spacecraft pose.

The application relates to the technical field of camera pose tracking, and discloses a low-cost camera six-degree-of-freedom tracking method, which comprises the following steps: acquiring three-dimensional coordinates of a Pivot in a virtual scene global coordinate system, a body coordinate system offset vector of the Pivot to a camera CMOS optical center, and an installation angle deviation matrix of an IMU relative to a camera body; and based on sensing data collected by the IMU, a pose rotation matrix in an IMU coordinate system is solved, and after installation angle deviation compensation and Pan zero point offset compensation are performed, the technical problem that in the prior art, when only relying on a low-cost inertial measurement unit (IMU), position integral error accumulation cannot be avoided, and after discarding visual collection and a high-precision mechanical encoder, it is difficult to balance hardware low cost, environmental adaptability and pose output stability is solved. The application provides a low-cost camera six-degree-of-freedom tracking method.

The application discloses a robot operation target 6D pose tracking method based on time sequence point cloud fusion. First, a point cloud video sequence of a robot operation scene is acquired and an operation target is segmented from the point cloud video sequence, then, dense corresponding relations among time sequence observation point clouds are established in a reverse prediction mode, and the dense corresponding relations are taken as a fusion criterion to perform point-by-point feature fusion on the time sequence observation point clouds, on the basis of the point-by-point feature fusion, a transformation pose between adjacent frames is predicted from the time sequence fused features in a confidence degree regression mode, finally, the transformation pose between the adjacent frames is continuously regressed to realize continuous pose tracking of the operation target. The application tracks a 6D pose of a robot operation target from a three-dimensional point cloud video sequence based on a deep learning technology, in view of irregularity and disorder of three-dimensional point cloud data, time sequence correlation information among observation point cloud data is fully utilized and effective feature fusion is performed, accurate and stable tracking is realized, and the application has good engineering practical value.

A pose tracking method and system, a mobile device, an electronic device and a storage medium. The pose tracking method includes: acquiring an image of a mobile device whereon is disposed with a light emitting unit for emitting signal light; based on the image, extracting, as a reference feature, a light spot feature corresponding to the light emitting unit on the image, and extracting a two-dimensional feature point corresponding to a three-dimensional feature point of the mobile device in the image; obtaining an initialization pose of the mobile device based on the two-dimensional feature point; and based on the initialization pose and the at least two reference features, optimizing the initialization pose for fine-tuning the initialization pose, to make the light spot feature corresponding to the initialization pose coincide with the reference feature. According to the embodiments of the present invention, the precision of pose tracking is improved, and the number of light emitting units required is reduced, such that a structure is simplified, and power consumption is reduced.

The application provides a three-dimensional rotation control method and system for radio frequency immunity test of an electric energy meter, which comprises the relative conversion relationship between a calibration camera and a three-dimensional turntable coordinate system, and establishes a 3D reference feature library of key points on the surface of the electric energy meter; in the test process, feature points are extracted from real-time monitoring images and matched with the reference library, and reliable 2D-3D point pairs are screened out by using a RANSAC algorithm; six degrees of freedom poses of the electric energy meter in the camera coordinate system are solved based on an EPnP algorithm, and then are mapped to the turntable coordinate system through coordinate transformation to obtain actual poses; the current actual poses and target test poses are respectively converted into the start and end angles of the axes of the turntable, and trajectory planning is performed in the joint space; the instantaneous poses are solved through real-time visual feedback, the pose deviation from the theoretical trajectory is calculated, and the angle compensation of each axis is iteratively solved by using a Gauss-Newton method, so that the control command is corrected, closed-loop pose tracking is realized, and the electric energy meter is stably brought to the predetermined test angle.

This invention discloses a deep learning-based method for endoscopic ultrasound probe recognition, posture tracking, and fusion display, belonging to the field of medical image processing technology. The method involves: acquiring endoscopic video frames as input to a posture detection model, outputting the probe frame and the coordinates and confidence scores of key points such as BlackCenter, Hole, and Top; after threshold filtering, calculating the center position and orientation angle based on the Hole-Top direction vector, and using confidence scores to drive adaptive hierarchical filtering to stabilize the angle and scaling; when there is occlusion or insufficient key points, calling position memory to maintain continuous tracking. The fusion end employs layered overlay, coordinate mapping, and fan-shaped mask clipping, combined with asynchronous inference and frame queue dropping to ensure real-time smoothness; it also provides automatic mask parameter estimation and calibration. This invention improves positioning and posture accuracy and anti-occlusion robustness, significantly suppresses jitter, reduces manual calibration costs, and enhances the consistency of fused display and the reliability of surgical navigation.

The application relates to the technical field of computer vision, in particular to a head posture tracking method and device, electronic equipment and a storage medium, wherein the method comprises the following steps: converting a video of a to-be-tracked target into an image sequence, inputting the image sequence into a pre-trained occlusion object recognition network, and outputting a mask image of an occlusion object in each frame of the image sequence; dividing the image sequence into a reference frame image and a candidate frame image sequence by using the mask image, and generating a to-be-completed frame image sequence by using the mask image of each frame of the candidate frame image sequence to cover the occlusion area; inputting the to-be-completed frame image sequence, the reference frame image and the mask image into a pre-trained deformation completion network, outputting a completed image sequence, and realizing head posture tracking of the to-be-tracked target based on the completed image sequence. Thus, the problems of false detection and jitter of the head posture caused by the poor robustness of the related art video-based head posture tracking method which is easily affected by the occlusion are solved.

The application discloses a visual real-time six-degree-of-freedom head pose estimation framework for a wearable optical pumped magnetometer-magnetoencephalography (OPM-MEG) system, aiming at solving the motion pose error problem caused by non-rigid head-helmet coupling. The framework includes two stages of initialization and real-time tracking: in the initialization stage, a hybrid differentiable rendering strategy is adopted, combined with soft rasterization and hard rendering technology, to realize high-precision initial pose alignment between the head and the helmet through multi-view contour matching; in the real-time tracking stage, based on the checkerboard marker, an improved PnP algorithm is used to estimate the real-time change of the head relative to the initial pose, and a multi-candidate solution screening and nonlinear optimization mechanism is introduced to suppress drift. The application does not need to rely on structured light or laser scanning hardware, can realize low-delay and high-robust real-time pose tracking, effectively reduces the registration error caused by motion, and improves the reliability and accuracy of the neural magnetic signal source positioning.

This invention discloses an improved DeepSort algorithm for pose tracking and a system, relating to the field of computer vision technology. The steps are as follows: acquiring the image to be detected; using the YOLOX model to perform target detection on the image to be detected to obtain candidate boxes, and then filtering them using a non-maximum suppression algorithm to obtain detection boxes; based on the results of the previous frame, using Kalman filtering to predict the target region of the current frame; calculating the Mahalanobis distance, apparent cosine distance, and pose cosine distance between the detection boxes and the predicted target regions; using the Hungarian algorithm and cascaded matching to compare the similarity of targets between two consecutive frames to perform target matching and obtain the tracking result; and correcting the tracking result using the target's pose information. This invention optimizes the DeepSORT algorithm, resulting in more accurate human tracking and pose information extraction, and solves the problem of poor target tracking performance caused by similar target appearances and inaccurate target regions.

A pose tracking device and method are disclosed. The device captures an image of one eye of a user. Based on the eye image, the device calculates the pose of the eye. The device captures a reflection image from the eye image. Based on the reflection image and the eye pose, the device calculates the pose of the user's head. The pose tracking method disclosed herein can calculate a more accurate head pose.

This invention discloses a robust 3D Gaussian sputtering SLAM method based on adaptive kernel smoothing, comprising: Step 1, capturing images of the target area to be constructed, and constructing an original 3D Gaussian map based on the previous t frames of RGB-D images; Step 2, performing CB-KNN-enhanced differentiable rendering on the original 3D Gaussian map; Step 3, performing iterative optimization of the original 3D Gaussian map using a CB-KNN-based SLAM system to complete map updates. The significant advantages of this invention are: First, extremely high rendering fidelity, closely matching the RGB image of the real scene, accurately restoring the details and colors of indoor furniture and decorations; Second, excellent depth accuracy, providing a reliable geometric basis for pose tracking and scene understanding; Third, outstanding robustness, consistently outputting high-quality reconstruction results even under complex indoor layouts and changing viewpoints.

Apparatuses, systems, and techniques to determined a pose of an object from a plurality of images. In at least one embodiment, the pose of an object is determined from at least two images of a video sequence using one or more neural networks, in which the neural network produces a distribution of pose information that is filtered to determine the current pose.

A system for determining the pose of an object. One or more cameras are arranged to provide images of an observed pose of the object from different perspectives of the object within a motion capture volume A segmentation process segments the images from each camera to provide an observed mask of the object from each image. A rendering process renders a digital twin of the motion capture volume and the object for each camera and generates rendered masks of the object. A mask alignment process aligns each observed mask with an associated rendered mask, thereby determining a rendered mask of the object that best matches an observed mask. The alignments are the basis of a transform process that orients the rendered object and its pose.

This invention relates to a 6-DoF pose estimation method for low-texture objects based on 3D Gaussian sputtering, belonging to the fields of computer vision and robotics. The steps are as follows: 1) Offline reference database construction: Acquire multi-view reference images of the low-texture object to obtain the object region. Train a 3D Gaussian model based on the object region and simultaneously construct a BOW initial pose retrieval library; 2) Online pose optimization: Acquire the query image and segment the object region. Assign an initial pose through BOW feature matching. Generate a rendered image using the 3D Gaussian model and the initial pose. Calculate the composite loss. Implement backpropagation of the loss gradient to the pose increment through differentiable rendering. Iteratively optimize the pose increment using the AdamW optimizer; 3) Continuous frame pose tracking: Design a Track mode for continuous frame scenes and output continuous frame tracking poses. This invention solves the problems of feature matching failure and insufficient pose estimation accuracy in low-texture scenes, and is suitable for practical applications such as industrial robot grasping and AR interaction.