65results about How to "Improve pose estimation accuracy" patented technology

This disclosure describes a method and system to perform object detection and 6D pose estimation. The system comprises a database of 3D models, a CNN-based object detector, multiview pose verification, and a hard example generator for CNN training.The accuracy of that detection and estimation can be iteratively improved by retraining the CNN with increasingly hard ground truth examples. The additional images are detected and annotated by an automatic process of pose estimation and verification.

The invention discloses a six-degree-of-freedom attitude estimation method based on Tag, By adding a tag to the object to aid detection, the Tag on the object is identified by the camera, Help SLAM complete initialization, After initialization, A feature point is continuously extracted from each frame of the image, and according to whether the speed matrix corresponding to the previous frame is empty, Camera pose estimation is carried out, and the initial value of camera pose estimation is taken as the initial value, and the re-projection error function of map points corresponding to feature points is adopted as the objective function to optimize camera pose, so as to obtain the optimized camera pose and map points corresponding to feature points, and then the camera pose is converted intothe pose of the object. The method of the invention has good robustness when the imaging quality is poor and the object moves at high speed, and has very high attitude estimation accuracy.

A method and system detects and localizes an object by first acquiring a frame of a three-dimensional (3D) scene with a sensor, and extracting features from the frame. The frame are segmented into segments, wherein each segment includes one or more features, and for each segment, searching an object map for a similar segment, and only if there is a similar segment in the object map, registering the segment in the frame with the similar segment to obtain a predicted pose of the object. The predicted poses are combined to obtain the pose of the object, which can be outputted.

The invention discloses an unmanned aerial vehicle pose estimation method based on visual inertia polarized light fusion, which adopts the combination of visual measurement and inertia measurement, and can well solve the problems of low output frequency, poor robustness and unobservable scale information of pure visual measurement after the visual measurement and the inertia measurement are fused.Meanwhile, the constraint among a plurality of images in visual measurement can also effectively inhibit integral drifting in inertial measurement and correct zero offset of inertial measurement. Onthe basis of vision measurement and inertia measurement, the course constraint of the polarized light compass is also added, the polarized light compass obtains the course information of itself by simulating the structure and function of sensing polarized light by desert ants. The method has the advantages of no accumulation of errors along with time and high stability, and can provide stable course constraint for a navigation sensor of an unmanned aerial vehicle. Compared with an existing visual inertia combination, the visual inertia polarized light fusion-based unmanned aerial vehicle poseestimation method provided by the invention can greatly improve the pose estimation precision of the unmanned aerial vehicle.

The invention discloses a spacecraft attitude estimation method based on norm-constrained cubature Kalman filter. The method comprises the following steps: a state equation of a spacecraft navigation system is established; a non-linear measurement equation of the spacecraft navigation system is established; a discretized state equation and a discretized measurement equation are obtained; a norm-constrained cubature Kalman filtering algorithm is used for the discretized state equation and the discretized measurement equation, and an attitude of the spacecraft is outputted. The norm-constrained cubature Kalman filter is used, norm constraint of quaternion for describing the spacecraft attitude is introduced into the attitude estimation, so that the problem of singular covariance existed in the process of attitude estimation without considering norm constraint is effectively solved, the influences of norm constraint conditions of quaternion on the spacecraft attitude estimation are effectively corrected by only adding small computational complexity, precision of attitude estimation is improved, navigation precision of the spacecraft is furthermore improved, and stability of the navigation process is enhanced.

The invention discloses a workpiece pose estimation method based on point cloud, which is mainly used for solving the problem of workpiece pose estimation in the field of automatic assembly, especially in the field of small part assembly. The method comprises the steps of point cloud data preprocessing, point cloud virtual view extraction and pose estimation. Wherein the pose estimation step further comprises the steps of performing iterative operation by adopting a particle filter algorithm based on a dynamic model of an iterative closest point algorithm, outputting the pose of the effectiveparticle in a weighted mean mode if an iterative stop condition is met, and calculating the pose of the target workpiece relative to the camera coordinate system through inversion operation. The invention also relates to an apparatus comprising a memory and a processor implementing the above method steps when the processor executes a program stored in the memory.

The invention provides a Sage-Husa adaptive unscented Kalman filter attitude data fusion method, belongs to the technical field of digital filter and multi-sensor data fusion and is mainly used for improving carrier attitude estimation precision. According to the method, error quaternion unscented Kalman filter is taken as a frame, data of a gyroscope, an accelerometer and a magnetometer are fused, through analysis of sensor errors, a relatively accurate sensor measurement model is established, and through combination of an error quaternion method and a UKF (Unscented Kalman Filter) filter algorithm, data fusion of the sensors is realized. The method is applicable to a nonlinear attitude measurement system and has relatively good anti-interference performance and attitude calculation precision. A carrier attitude calculation problem under a complex environment can be effectively solved.

The invention discloses a vector constraint-based random characteristic point selection method for landing position detection. The method comprises the steps of 1, performing acquisition and synchronous uploading of a landing region image; and 2, processing the landing region image: 201, processing the landing region image at an initial sampling moment; 202, processing the landing region image at the next sampling moment by Harris corner extraction, Harris corner matching, SIFT characteristic extraction, SIFT characteristic point matching, characteristic point fusion, characteristic point combination generation and optimal characteristic point combination screening; and 203, returning to the step 202 of processing the landing region image at the next sampling moment. The method is simple in step, reasonable in design, convenient to realize, high in practicality and good in using effect; characteristic points for pose estimation can be simply, conveniently and quickly selected from characteristic points extracted from unknown landing region images; and the pose estimation precision can be effectively improved.

The invention belongs to the technical field of aircraft attitude estimation by use of a robustness filtering technology, and relates to a robustness recursion filtering method for aircraft attitude estimation under the condition of measurement interference. The method comprises: acquiring the output data of a gyro and a star sensor during the motion process of an aircraft; establishing an aircraft non-linear state space model based on attitude quaternion under the condition of measurement interference; updating time, and solving a step state prediction value and an upper bound of a prediction variance; performing robustness recursion filtering measurement updating, solving an optimum filtering gain, accordingly solving a state estimation value and the upper bound of the variance at k+1 time, and performing forced normalization constraining on a quaternion portion in state estimation at the k+1 time; and outputting the attitude quaternion and a gyro drifting result, and finishing attitude estimation. According to the invention, a robustness filtering design based on a minimum variance is employed, such that an optimum filtering gain design under the condition of minimum variance significance can be realized, and the attitude estimation precision of a system can be improved.

The present invention discloses a combination navigation method, which comprises: acquiring an attitude measurement value in a first speed interval range by using a first combination filtering, wherein the first combination filtering is the INS / GNSS combination filtering with an observation equation containing the INS and GNSS attitude angle measurement difference; acquiring an attitude measurement value in a second speed interval range by using a second combination filtering and an added attitude filtering, wherein the second combination filtering is the INS / GNSS combination filtering being not contain the INS and GNSS attitude angle measurement difference, and the added attitude filtering is used for carrying out attitude filtering on the filtering result of the second combination filtering; and acquiring an attitude measurement value in a third speed interval range by using the second combination filtering. The invention further discloses a combination navigation system. With the method and the system of the present invention, the attitude observability and the estimation precision are improved.

The invention discloses a mobile robot pose estimation method and system based on multi-sensor tight coupling, and belongs to the technical field of robot localization, and the method comprises the steps of carrying out the feature matching of a current frame RGB-D image collected by a camera and a previous frame RGB-D image, and calculating a visual re-projection error in the matching process; integrating the data measured by the IMU, and constructing an IMU integral residual error; extracting edge feature points and plane feature points from the point cloud collected by the laser radar, calculating the distance from the edge feature points to the edge line and the distance from the plane feature points to the plane, and constructing a laser point cloud geometric residual error; performing pose estimation by taking the minimum visual re-projection error, IMU integral residual error and laser point cloud geometric residual error as targets to obtain a local pose; and updating the laser point cloud map by using the local pose, and performing global optimization on the laser point cloud map to obtain a global pose. According to the invention, the positioning precision and robustness of the mobile robot in complex motion and complex environment are improved.

The invention discloses an unsupervised pose and depth calculation method and a system. The unsupervised pose and depth calculation method mainly adopts the following modules: a pose prediction network model TNet, a depth estimation network model DMNet, a visual reconstruction model V and an error loss function module. The method comprises the steps of calculating a forward motion relative pose and a backward motion relative pose; calculating the depth estimation result of the image and the corresponding depth of the image, summing the reconstruction error, the smoothing error and the twin consistency error to obtain a loss function, performing iterative updating until the loss function converges, and finally calculating the relative pose and the predicted depth map of the camera accordingto the trained model Tnet and model DNet.

The invention discloses a method for determining a relative attitude of a microsatellite by using a dual array antenna. A relative attitude dynamics equation is used as a system equation to form a probe frame, and then a transmitting and receiving timing sequence is designed; two array antennas of a main spacecraft receive a transmitted signal from a slave spacecraft array antenna which is used asan output signal, and moving correlation is carried out according to the output signal and a probe signal, so as to obtain an array impulse response matrix; two-dimensional spectral peak search is performed to obtain two sets of wave leaving angles and wave arrival angles; four unit line-of-sight vectors are obtained by using the two sets of wave leaving angles and wave arrival angles; a measurement equation is constructed by using the bidirectional unit line-of-sight vectors and a relative attitude rotation matrix to obtain a measurement noise variance matrix; the TRIAD algorithm is used toobtain an attitude quaternion at the initial moment; and then the relative attitude and a relative rotational angular velocity which are convergent gradually are obtained according to the system equation and the measurement equation. The method for determining the relative attitude of the microsatellite by using the dual array antenna has low computational complexity, high estimation precision andwide application range.

The invention discloses a medicine box detection method based on three-dimensional point cloud and image data fusion, and the method comprises the steps: inputting a medicine box image collected by a camera into an optimized U-shaped full convolutional neural network, and extracting a medicine box feature image; in the feature extraction part, a grouping residual error convolution module is used for extracting a preliminary feature image, a cavity space convolution pooling pyramid module is used for extracting feature image information of different scales of the preliminary feature image, and a mixed attention module is used for fusing the feature image information of different scales to obtain a two-dimensional fusion feature image; a segmented medicine box image is obtained through up-sampling; and whether the segmented medicine box image meets the detection requirement is judged, if not, three-dimensional information of the medicine box is extracted, a target in the image is positioned through a two-dimensional target detection network, and a cone point cloud corresponding to the two-dimensional detection frame is obtained according to a camera geometric imaging model. A Point Net point cloud network and a feature fusion network layer are adopted to carry out instance segmentation on the cone point cloud to obtain all target points. A target centroid is estimated by using a T-Net network, a target point cloud is moved to a centroid coordinate system, then estimation of parameters of a three-dimensional bounding box is obtained through a parameter estimation network and a feature fusion network layer, finally, the size and orientation of a medicine box are obtained, the type of the medicine box is judged, and the character 0 of the medicine box is recognized in combination with image information. The problems of long time consumption, high false detection rate and the like in existing medicine box detection are solved.

The invention discloses an adaptive combination navigation antenna beam control method for a satellite communication in motion communication system. The method comprises the following steps: (1) building a system equation of the satellite communication in motion system, calculating the state variable x of a combination navigation attitude estimation system according to the position error amount deltar of a carrier, the speed error amount deltav of the carrier and the attitude angle error amount Psi of the carrier, and meanwhile setting the observed quantity of a satellite communication in motion combination navigation attitude estimation algorithm with a main controller; (2) building a corresponding satellite communication in motion combination navigation measurement equation according to different driving environments of the carrier, and solving a measurement equation coefficient matrix Hk; and (3) performing attitude computation through adaptive extended Kalman filter to obtain the three-dimensional attitude angle of the carrier, and controlling three motors to adjust the direction of antenna beams according to the three-dimensional attitude angle of the carrier in order that an antenna is aligned with a satellite. Through implementation of the method, accurate alignment between the antenna and the satellite can be realized, and the cost is low.

The invention discloses a method for estimating attitude and parameters of a non-cooperative spacecraft without a gyroscope, and the method comprises the following steps of considering the problem that the relative coordinate system of the target non-cooperative spacecraft established according to the optical sensor does not coincide with its body coordinate system, establishing an attitude kinematics and dynamics model of the non-cooperative spacecraft of the quick-roll space in the relative coordinate system of the target non-cooperative spacecraft; selecting the error of the attitude quaternion vector part, the three-axis rotational angular velocity and the rotary inertia ratio of the target non-cooperative spacecraft as the filtering variable, and deducing the discrete state equationsand observation equations of the linearized system; designing a multiplicative extended Kalman filter, and improving the five-step prediction to improve the accuracy of filtering estimation and the fastness of follow-up, thereby solving the problem of increased filtering errors caused by over-low sampling frequency of the optical sensor.

The invention discloses a quaternion-based attitude estimation method. The quaternion-based attitude estimation method comprises the following steps: constructing a Kalman filter according to quaternions; acquiring observed quantities of a magnetometer and an accelerometer; determining whether the observed quantity of the accelerometer exceeds the limit or not, and if the observed quantity of theaccelerometer and the observed quantity of the magnetometer do not exceed the limit, updating the observed value of the Kalman filter to obtain an updated quaternion; and performing normalization andcovariance matrix symmetry processing on the updated quaternion to obtain final attitude information. The invention further discloses electronic equipment and a computer readable storage medium. According to the quaternion-based attitude estimation method, all-attitude estimation can be performed without linearizing the quaternion, and the quaternion-based attitude estimation method is insensitiveto the initial attitude of a carrier; observation results of various sensors are fused, zero offset existing in the working condition of a gyroscope is considered, modeling is carried out on noises of the zero offset, so that the attitude estimation precision, accuracy and robustness are improved.

The invention discloses a spatial non-cooperative target pose estimation method based on deep learning, and the method comprises the steps: 1, dividing pose category intervals, generating spatial non-cooperative target image data, and marking a pose category label, a pose numerical value label and a position label, and obtaining a marking data set of a spatial non-cooperative target, which comprises a training set, a test set and a verification set; 2, constructing a neural network applied to spatial non-cooperative target pose estimation based on an AlexNet network, removing full connection layers at the tail end of the network, and then connecting the four full connection layers in parallel; 3, designing loss functions of four branches; 4, inputting the training set and the verification set into the constructed neural network, training the network by utilizing a designed loss function, and storing a neural network model when the loss function is converged to a global minimum value; and 5, performing pose estimation on the spatial non-cooperative target by using the trained neural network model. According to the invention, pose estimation of a spatial non-cooperative target can be realized through a single camera and a single image.

The invention discloses a pose calculation method combining deep learning and geometric optimization. The system comprises a pose estimation module PoseNet and a depth estimation module DepthNet, a pose estimation method based on deep learning and an optimization strategy based on geometric constraints are integrated into a visual odometer frame, luminosity errors, feature point reprojection errors, adjacent inter-frame constraints and constraints in a local map formed by a section of continuous frames are calculated, and real-time and accurate pose estimation can be carried out.

An embodiment of the invention discloses a method for determining vehicle pose, a vehicle-mounted system and a vehicle, which obtain a fused pose of the vehicle according to pose output fusion of monocular systems, thereby improving the pose estimation precision of the vehicle. The method of the embodiment of the invention comprises the steps of: acquiring first inertial navigation information andfirst vehicle odometer information of the vehicle; determining a first current estimated pose of the vehicle according to the first inertial navigation information and the first vehicle odometer information; acquiring N monocular camera external parameters of the vehicle, wherein N is an integer greater than 1; integrating an inertial navigation mileage calculation method and a visual SLAM algorithm of each monocular camera according to the N monocular camera external parameters and the first current estimated pose, so as to estimate current poses of the N monocular cameras; performing coordinate transformation on the current poses of the N monocular cameras respectively to obtain N converted poses of the vehicle; and acquiring the fused pose of the vehicle according to the N converted poses.

The invention belongs to the field of underwater geomagnetic assisted navigation, and particularly relates to an attitude estimation method and device for a multiplicative error quaternion based geomagnetic tensor field assisted gyroscope. A state equation of the attitude estimation device is established by error quaternion, gyroscope scale factor error and gyroscope scope constant bias error, andvector magnetometer groups in strapdown to a carrier are used for measuring five independent components of a geomagnetic tensor field under the carrier system; measurement values of the five independent components are subjected to subtraction with measured values of the independent components of the geomagnetic tensor field under the carrier system, and the independent component difference of thegeomagnetic tensor field is obtained; a Kalman filter is used for estimation and compensation, and estimated values of the attitude quaternion, gyroscope scale factor and constant bias are obtained;attitude of the carrier is determined by the attitude quaternion. The problem of singularity of the state covariance matrix is solved by using multiplicative error quaternion as the attitude estimation variable; the influence of vector direction of the geomagnetic background field on attitude estimator system observation is reduced by using the difference of the independent components of the geomagnetic tensor field as observed quantity.

The invention provides a pan-tilt attitude estimation method based on extended Kalman filtering. The method employs quaternion to represent the current attitude of an object, a system state quantity comprises the offset error of quaternion and angle increment, an accelerometer and a magnetometer are used to correct the offset error of angle increment, and the attitude estimation is be more accurate. Accelerometer correction and magnetometer correction are implemented in two stages so that the accelerometer correction and the magnetometer correction do not interfere with each other, and the attitude estimation accuracy is improved; and in the correction of the accelerometer, the third vector of the quaternion in the correction is set to be zero, and in the correction of the magnetometer, the first vector and the second vector of the quaternion in the correction are set to be zero. According to the invention, more accurate attitude estimation information can be obtained.

The invention discloses a measurement delay filtering method for a small field star sensor, and belongs to the technical field of aircraft attitude estimation. In view of the conditions that multiplicative noise exists in a non-linear attitude estimation system state model and unknown interference exists in a measurement model, an aircraft attitude estimation model with the existence of the indetermination of the two models is established, meanwhile, the upper boundary ranges of prediction variance and estimation variance are found by using a robust algorithm design based on an expansion Kalman filtering structure, and then optimal filtering gain is designed by using a minimum variance theoretical design. The aircraft attitude is determined by using expansion Kalman filtering, the truth isminimum variance criterion estimation which is different from that of general frequency domain filters, the method can be used for processing attitude estimation under undetermined conditions, the method has the advantages of high instantaneity, and low dependence degree on prior information, the problem that delay exists in the star sensor is solved, and relatively high attitude estimation accuracy can also be guaranteed even if in information uncertainty conditions.

The invention relates to a small unmanned aerial vehicle MARG heading and attitude estimation method based on an adaptive EKF algorithm, and belongs to the technical field of integrated navigation. According to the method, firstly, measurement updating of a magnetometer and an acceleration meter are carried out separately, magnetometer error evaluation is added, when the magnetometer error is large, the magnetometer error can be isolated, and interference on horizontal attitude estimation is avoided; secondly, the invention further provides an adaptive filtering algorithm for the external acceleration, in order to improve the estimation precision, a triaxial acceleration module value is not directly judged, a triaxial residual error is analyzed, and then corresponding measurement noise issubjected to self-adaptive adjustment, so that useful acceleration information is prevented from being lost, and the attitude estimation precision is improved.

The invention provides a dynamic residual threshold adaptive quaternion particle filter attitude calculation data fusion method, which belongs to the technical field of digital filtering and multi-sensor data fusion, and is mainly used for improving the precision of carrier attitude estimation. According to the method, a standard particle filter algorithm is used as a framework, data of a gyroscope, an accelerometer and a magnetometer are fused, a relatively accurate sensor measurement model is established by analyzing sensor errors, quaternion is selected as an attitude parameter, a reasonable range of residual errors is predicted and judged, then a measurement noise matrix of a system is adaptively adjusted, and data fusion of sensors is realized. The method is suitable for a nonlinear attitude measurement system, has good anti-interference performance and attitude calculation precision, and can effectively solve the problem of carrier attitude calculation in a complex environment.

The invention provides a robot operation pose control method based on visual-touch multi-scale positioning. The robot operation pose control method comprises the steps that firstly, the operation environment and the state of a to-be-operated target object are measured and obtained through a visual sensor, and according to collected image information and the safety distance, the estimated pose of the to-be-operated target object and the estimated target pose of a mechanical arm system are preliminarily obtained; then, the relative pose deviation of the tail end of the mechanical arm system andthe to-be-operated target object is measured and obtained through a touch sensor; and finally, according to the robot operation precision requirement, the estimated target pose of the mechanical arm system is iteratively controlled and adjusted. Through the visual-touch multi-scale positioning method, local information is increased through touch sensing, and the robot operation pose is precisely controlled. The invention provides a pose error measuring method based on the touch sensor, the relative pose deviation of the tail end of the mechanical arm system and the to-be-operated target objectis precisely measured in a six-dimensional error space, and the robot operation pose error measuring precision is improved.

The invention discloses a visual inertial integrated navigation method fusing semantic features. The method comprises steps of RGBD visual sensor data S(k), accelerometer data and gyroscope data at the k moment being collected, and the current pose T (k) of a camera being obtained through calculation by using a visual odometer according to the visual sensor data S (k); performing semantic plane feature extraction and matching between two adjacent image frames by using the visual sensor data S (k); performing pre-integration between two adjacent image frames by using inertial sensor data; combining the semantic plane observation residual error, the visual odometer relative pose observation residual error and the inertia pre-integration residual error to optimize and solve the carrier navigation information; and outputting the carrier navigation information and camera internal parameters. According to the method, positioning accuracy and robustness of the navigation system can be effectively improved.

This disclosure describes a method and system to perform object detection and 6D pose estimation. The system comprises a database of 3D models, a CNN-based object detector, multiview pose verification, and a hard example generator for CNN training.The accuracy of that detection and estimation can be iteratively improved by retraining the CNN with increasingly hard ground truth examples. The additional images are detected and annotated by an automatic process of pose estimation and verification.

The invention relates to an iterative 6D pose estimation method and device based on deep learning, and the method comprises the steps: carrying out the coding of the output of a 3D-2D key point corresponding relation extraction network, and coding the output of the 3D-2D key point corresponding relation extraction network into an input format of a pose coarse estimation network; the method comprises the following steps that: firstly, object 6D postures are input into a pose coarse estimation network constructed by utilizing an MLP, a pooling layer and a full connection layer, so that the pose coarse estimation network is combined into an integral network, the object 6D postures can be directly output, an end-to-end 6D object pose estimation network is formed, and a loss function of the 6D object pose estimation network is a function capable of directly reflecting and resolving object 6D pose parameters; and optimizing 6D attitude parameters output by the 6D object pose estimation network by using an orthogonal iterative algorithm. According to the method, the problems of long time consumption and poor repeatability of calculating the target attitude based on the PnP algorithm when many abnormal values exist are solved, and the attitude estimation efficiency and the attitude estimation robustness and accuracy are improved.