182 results about "Graph optimization" patented technology

Disclosed in the invention is an improved method of a RGB-D-based simultaneously localization and mapping (SLAM) algorithm. The method comprises two parts: a front-end part and a rear-end part. The front-end part is as follows: feature detection and descriptor extraction, feature matching, motion conversion estimation, and motion conversion optimization. And the rear-end part is as follows: a 6-D motion conversion relation initialization pose graph obtained by the front-end part is used for carrying out closed-loop detection to add a closed-loop constraint condition; a non-linear error function optimization method is used for carrying out pose graph optimization to obtain a global optimal camera pose and a camera motion track; and three-dimensional environment reconstruction is carried out. According to the invention, the feature detection and descriptor extraction are carried out by using an ORB method and feature points with illegal depth information are filtered; bidirectional feature matching is carried out by using a FLANN-based KNN method and a matching result is optimized by using homography matrix conversion; a precise inliners matching point pair is obtained by using an improved RANSAC motion conversion estimation method; and the speed and precision of point cloud registration are improved by using a GICP-based motion conversion optimization method.

The invention relates to a visual inertial navigation SLAM method based on ground plane hypothesis. According to the method, feature points are extracted from an image to perform IMU pre-integration,a camera projection model is established, and camera internal parameter calibration and external parameter calibration between an IMU and a camera are performed; a system is initialized, a visually observed point cloud and a camera pose are aligned to the IMU pre-integration, and a ground equation and the camera pose are restored; the ground is initialized to obtain a ground equation, the ground equation under the current camera pose is determined and back projected to an image coordinate system, and a more accurate ground region is acquired; and based on state estimation, all sensor observation models are derived, camera observation, IMU observation and ground feature observation are fused to do state estimation, a graph optimization model is used to do state estimation, and a sparse graph optimization and gradient descent method is used to realize overall optimization. Compared with previous algorithms, the precision of the method is greatly improved, estimation of the camera pose can be limited globally, and therefore accuracy is greatly improved.

The invention provides a neural network compiler architecture and method. The compiler architecture comprises: a calculation graph construction module which is used for constructing a universal firstintermediate representation based on inputted different types of model files, wherein the first intermediate representation is in a graph form; a calculation graph optimization module which is used for carrying out graph optimization on the first intermediate representation to obtain a second intermediate representation in a graph form; and an instruction generation module which is used for carrying out scheduling optimization on the second intermediate representation to obtain a fine-grained third intermediate representation, and compiling the third intermediate representation into an instruction code executed on the hardware platform based on the hardware platform. The modules in the compiler architecture are matched with various intermediate representations with different granularitiesand attributes, so that various deep learning frameworks and rear-end hardware platforms can be handled with extremely high expandability and compatibility, and efficient and accurate code optimization capability is provided.

The invention discloses a VSLAM method based on a multi-characteristic visual odometer and a graph optimization model, and belongs to the robot SLAM field; the method comprises the following steps: firstly using a FAST (features from accelerated segment test) and an improved LSD algorithm to extract point and line features in a color image; further using different descriptors to describe characteristics; then carrying out characteristic matching; finally using an improved NICP (normal iterative closest point) algorithm and a PnL (perspective n line) algorithm to estimate a robot initial posture. The method can extract image line features so as to enlarge algorithm application scenes, can obtain a well robot initial posture, uses a Bayes network to express a multi-characteristic visual odometer, obtains a factor graph on the basis of the Bayes network, uses a maximum posterior probability to estimate a robot global posture in the factor graph, and uses a Gauss-Newton method to solve themaximum posterior probability to obtain the updated posture graph; finally, the posture graph and three dimensional points of corresponding frames are fused to obtain a reconstructed three dimensional map.

The invention relates to an Android malicious software sorting method based on a dynamic behavior dependency graph. The sorting method includes the steps that an APP is run through a user-defined Dalvik virtual machine, and dynamic behavior information such as framework layer interface calling behaviors and dependency among the behaviors is extracted; the corresponding dynamic behavior dependency graph is constructed according to the dynamic behavior information; the dynamic behavior dependency graph is optimized and divided into subgraphs; similar subgraph structures are extracted from a set composed of Android malicious software of different types and are used as essential characteristics; according to the essential characteristics, model training is conducted on a training set composed of known malicious software and normal software to obtain a classifier; unknown APPs are classified and judged through the classifier; the method is verified and assessed. Similarity of the behavior subgraphs is measured with the graph editing distance, basic characteristics are found on the basis, and the sorting method has good flexibility and expandability.

The invention provides a tracking method of a relative movement angle and an angular velocity of a slowly rotating space fragment. The method comprises the following steps: after image information of a slowly rotating space fragment object is acquired, performing extraction, matching, tracking and pose solving of features points on current frames; performing object local space point three-dimensional reconstruction; performing object global closed-loop detection to eliminate accumulative errors; and obtaining an angular velocity and a total angle of object movement, and eliminating interframe revolution errors. According to the invention, from the perspective of image features, by use of a graph optimization method, tracking measurement of the slowly rotating space fragment is realized through a feature relay between key frames and key points, at the same time, the accumulative errors are eliminated by use of a closed-loop detection means, and thus the measurement precision is improved.

The invention is suitable for the technical field of computers, and provides a method and a device for indoor navigation of an unmanned aerial vehicle, the unmanned aerial vehicle and a storage medium. The method includes the following steps: photographing the current scene through a monocular camera arranged on the unmanned aerial vehicle to obtain corresponding scene images, acquiring the inertial measurement data of the unmanned aerial vehicle through an inertial measurement unit arranged on the unmanned aerial vehicle, preprocessing the inertial measurement data, combining tracked scene image characteristics with the preprocessed inertial measurement data to generate fusion data, nonlinearly optimizing every frame of the data, and carrying out closed loop detection and attitude graph optimization on the state of the unmanned aerial vehicle according to the fusion data and every frame of the nonlinearly optimized scene images in order to realize the navigation of the unmanned aerial vehicle. The method and the device have the advantages of realization of the automatic navigation of the unmanned aerial vehicle in the indoor environment with weak or no GPS signals, effective reduction of the navigation cost of the unmanned aerial vehicle, and improvement of the navigation accuracy and the precision efficiency of the unmanned aerial vehicle.

The invention discloses a 360-degree three-dimensional reconstruction optimization method based on continuous phase dense matching. According to the method, 360-degree reconstruction of the three-dimensional point cloud of the measured object can be realized rapidly; a reconstruction result is subjected to nonlinear optimization; the method is realized through the following scheme: the method comprises the following steps: firstly, carrying out calibration of a digital projector and a camera; obtaining a corresponding structured light deformation image; calculating the phase order of the deformed fringe pixel points, and determining the epipolar lines of the deformed fringe pixel points in different camera imaging planes of the camera array at the same time, thereby establishing epipolar geometry and equiphase joint constraints, calculating the dense matching of structured light images at different viewing angles, and generating a phase dense matching relationship of the deformed fringe pixels at different angles; initializing a camera transformation matrix and a three-dimensional point cloud initial point by utilizing a phase dense matching relationship and a triangularization principle, constructing an objective function and a graph optimization model thereof, and solving the objective function and the graph optimization model; and performing triangulation surface reconstruction on the optimized three-dimensional point cloud to obtain a complete 360-degree three-dimensional target reconstruction model of the measured target.

The invention discloses a multi-binocular camera motion calibration method based on a non-common-view area, and relates to the field of mobile robot vision SLAM. A device comprises two binocular cameras, a camera plane calibration plate and a control computer. The method is divided into three parts, firstly, the Zhang Zhengyou calibration method is used for obtaining the internal parameters, the external parameters and the distortion coefficients of two to-be-tested cameras, then the binocular camera conversion relation is calculated according to the common viewpoint seen after the binocular camera moves, and finally a graph optimization theory is used for optimizing a conversion matrix to finally obtain a high-precision calibration result. The method not only can be used for calibration calculation of the multi-binocular camera in the visual SLAM field, but also can be used in other visual measurement fields, is not limited to the number of the binocular cameras, and has the advantages of being simple in device, flexible and convenient to use, good in robustness, accurate and reliable in detection result and wide in application range.

The invention belongs to the technical field of unmanned aerial vehicle navigation indoor positioning, and discloses an unmanned aerial vehicle vision-inertia fusion indoor positioning method. The method comprises steps of obtaining scale information of the feature points by using a binocular parallax method; tracking the feature points by using a pyramid LK optical flow algorithm; removing the feature points which are mismatched; pre-integrating inertial measurement data, and keeping data synchronization between sensors; and constructing a sliding window, carrying out selection and marginalization management on the key frame, and estimating the position, attitude, speed and gravity direction of the unmanned aerial vehicle and the drift of an inertial device by adopting a graph optimization mode. Finally, the functions of indoor positioning of the unmanned aerial vehicle in the dynamic environment and closed-loop detection and re-indoor positioning of the reached position are designed,and the precision and robustness of indoor positioning of the unmanned aerial vehicle are effectively improved.

A traditional laser triangulation algorithm is high in location precision, and it is likely to be influenced by blocking of an obstacle to cause the phenomenon that AGV cannot be accurately located; the graph optimization SLAM algorithm ensures the precision and accuracy of map building through the method for building a map through front-end map building and rear-end optimization, and the integrity of map building is achieved through a closed-loop detection method. In the environment such as a garage which is less in feature, single in feature and symmetrical in feature height, the closed-loopdetection function cannot be well achieved, the phenomenon that the built map has multiple areas of overlap is caused, and the map cannot provide navigation and location. Accordingly, the method forconducting optimization by combining the laser triangulation algorithm with the SLAM algorithm is put forward, a reflector arranged in the environment in advance divides a large environment into countless small environments, and the small environments are numbered independently. The location can be conducted through laser triangulation in the small environments, after the environments are numberedin a regional mode, each area at which the AGV arrives can be detected, different areas have different numbers, and finally a graph optimization SLAM closed-loop detection signal is formed; in this way, the built map is high in precision.

The invention provides a water supply-power generation-ecological multi-objective scheduling graph optimization method based on ecological flow. The method comprises the following steps: 1, collectingreservoir long-series reservoir inflow runoff data, water demand data of a water supply area, downstream river channel ecological base flow data and an existing conventional scheduling graph; 2, determining a downstream river long-series ecological flow threshold interval by adopting a month-by-month frequency method; 3, constructing a water supply-power generation-ecological multi-objective scheduling graph optimization model; and 4, solving the optimization model by adopting an 'optimization-simulation' technology to obtain an optimization scheduling graph, and determining various benefit indexes and a reservoir operation process. According to the invention, the water supply-power generation-ecological multi-target scheduling graph optimization method combining the ecological flow threshold interval is proposed for the first time. The method can guarantee that the power generation flow, the discharge flow and the water supply amount are optimal scheduling results after power generation, ecology and water supply benefits are comprehensively considered, and a new thought is provided for achieving balanced coordination among multiple targets of reservoir water supply, irrigation, ecology and power generation.

The invention discloses a mapping method and device based on a multi-line laser radar, a medium and equipment. The mapping method comprises the steps: projecting three-dimensional point cloud data collected by the multi-line laser radar to the horizontal plane where the laser radar is located to generate two-dimensional laser radar point cloud data; performing inter-frame matching on the generatedtwo-dimensional laser radar point cloud data to realize data association; calculating a multi-line laser radar pose according to the position of the point cloud obtained by matching on the map, and managing a multi-line laser radar coordinate sequence by adopting a graph optimization mode to obtain a motion trail pose graph representing the robot; and generating an occupied grid map according tothe multi-line laser radar pose, the two-dimensional laser radar point cloud data and the counting weight corresponding to the distance between the obstacle and the laser radar. According to the method, the problems that the low-height obstacle is close to the mapping path and a measurement blind area exists and cannot be modeled as an obstacle are solved, the modeling of the low-height obstacle in the map can be well processed, and the mapping accuracy and reliability are improved.

Disclosed a multi-station scanning global point cloud registration method based on graph optimization, including acquiring multi-station original three-dimensional point cloud data; based on initial registration of targets, completing initial registration of point cloud data at adjacent stations by virtue of the target at each angle of view; calculating a point cloud overlap area at adjacent angles of view, and calculating areas of overlap regions of adjacent point cloud by a gridded sampling method; constructing a fine registration graph structure, and constructing a fine registration graph by taking point cloud data of each station as a node of the graph and taking an overlap area of the point cloud data of adjacent stations as a side of adjacent nodes of the graph structure; and based on loop closure fine registration based on graph optimization, gradually completing point cloud fine registration of the whole aircraft according to a specific closure sequence.

The invention discloses a method suitable for indoor plane 2D SLAM mapping of a building. The method comprises the following steps that: firstly, a high-precision point cloud generated by the laser radar is used to extract a scanning key frame by using an improved graph optimization algorithm, After optimization of the scanning key frame, the scanning key frame is inserted into a sub-image; secondly, the optimized scan frame will be stored in a background for standby application in closed-loop detection; thirdly, through global optimization, the cumulative error in the SLAM process is reduced;and finally, the scanning frame stored in the background is transferred to perform closed-loop detection, and when the closed-loop is achieved, an accurate two-dimensional plan can be obtained. The mapping method provided by the invention is more accurate and much simpler than the traditional indoor mapping method, and the map provided by the laser radar can be used for robot navigation. The method can comprehensively and accurately construct the indoor plane environment so as to have great significance to the development of surveying and mapping disciplines.

the invention discloses an indoor self-mapping method for storage environment. The indoor self-mapping method for the storage environment comprises the following steps that environment information iscollected by installing various sensors on a storage forklift, data storage, data fusion and data processing are carried out, and the real-time creation of a grid map is realized through a SLAM algorithm based on the graph optimization; all exit areas on the currently established grid map are determined, the exit areas are sequenced based on a distance evaluation function value, and an exit area with a largest value is selected as a next exploration target; if the next exploration target is not reached within specified time, and then a sub-large value is selected as the next exploration target; and after the next exploration target is reached, a same method is adopted for creating and expanding the grid map. According to the indoor self-mapping method for the storage environment, the identification of small obstacles can be enhanced, further obstacle information is additionally arranged on to the map, and the efficiency of mapping and the accuracy of the map are improved.

The invention discloses a feature matching method based on depth re-projection and space consistency, and the method comprises the following steps: obtaining an RGBD image and a depth image through anRGBD camera, generating a three-dimensional point cloud, and extracting a plane through RANSAC; performing feature point extraction on the generated point cloud image by using ORB, and establishing adescriptor; establishing a rough corresponding relation between the two frames through a KNN algorithm; obtaining a more reliable corresponding relation by utilizing the space consistency of featurepoints, carrying out feature matching, giving 3D coordinates, and obtaining reliable matching features through graph optimization. According to the method, the corresponding relationship between the frames is extracted more reliably by utilizing the spatial consistency of the feature points, so that the accuracy and robustness of feature matching are improved.

The invention discloses a method for generating an attack path and an attack graph based on multiple dimensions, and belongs to the technical field of network security. According to the method, vulnerability entry samples are collected, a vulnerability knowledge base is established, and a front permission and a rear permission are marked for each vulnerability entry sample; features, including vulnerability description text features and CVSS index features, of each vulnerability entry sample in the vulnerability knowledge base are extracted; text preprocessing is performed on the vulnerabilitydescription text features; an attack mode is defined by using a triple < the prepositive authority, the postpositive authority and the vulnerability entry >; an attack mode knowledge base is constructed; for the target industrial control network, an attack graph is generated by adopting a breadth-first forward attack graph generation algorithm based on an attack graph optimization strategy. The multi-dimensional attack paths and the high-dimensional attack graphs corresponding to various attack modes are generated by analyzing the environment attributes of different terminals, the attack graph generation process is optimized, and the attack graph generation efficiency is improved.

The invention belongs to the computer vision field and especially relates to an RGB-D visual odometer based on GMS characteristic matching and sliding window pose graph optimization. A GMS (grid motion statistics) algorithm is used to replace a commonly used distance threshold + RANSAC (random sample consistency) algorithm in the prior art to carry out mismatching rejection. When relative motion between images is large and brightness changes greatly, a sufficient number of correct matching point pairs can still be screened, which improves robustness of the system. A sliding window pose graph optimization technology is used to reduce a cumulative error of pose estimation. Compared with the prior art scheme in which a local map is maintained or a more complex objective function is designed,by using the odometer of the invention, higher real-time performance is possessed and at the same time accuracy of the visual odometer can be ensured.

The invention discloses an image positioning method and device based on ray model three-dimensional reconstruction. The method comprises the steps of acquiring a plurality of images in a plurality of scenes in advance, and extracting features of the images to obtain a plurality of feature point sets; carrying out feature matching of every two images for the plurality of images, generating a corresponding eigenmatrix according to the feature matching, and carrying out noise processing therefor; carrying out three-dimensional reconstruction based on a ray model according to the feature matching and the eigenmatrix after the noise processing, to generate a three-dimensional feature point cloud and a reconstructed camera posture set; obtaining a query image, and extracting features thereof to obtain a corresponding two-dimensional feature point set; and carrying out image positioning according to the two-dimensional feature point set, the three-dimensional feature point cloud and the reconstructed camera posture set based on a positioning posture graph optimization frame. The method improves the reconstruction effect, reduces the acquisition cost in the reconstruction process, increases the calculation speed, and improves the image positioning precision in the image positioning process.

The invention provides a large-scale three-dimensional environmental map establishing method based on the graph optimization theory. The large-scale three-dimensional environmental map establishing method based on the graph optimization theory comprises the steps that firstly, local poses of a mobile robot are sequentially estimated by means of a clipping iterative closest point algorithm, the reliability of an estimation result is judged, if the result is not reliable, the local posts of the mobile robot at the current moment are re-estimated by means of a rapid corresponding propagation algorithm based on multi-scale descriptors, a pose graph is gradually established at the same time, the peaks of the graph express the pose at each moment of the mobile robot, and edges of the graph express the constraint between connected poses; then, a closed loop hypothesis and a verification method are provided and used for detecting the closed loop of the pose graph; and finally, a pose constraint equation is solved by means of a motion averaging method, and then the accurate overall poses of the mobile robot are obtained. An experiment result shows that the large-scale three-dimensional environmental map establishing method based on the graph optimization theory can well establish a large-scale 3D environmental map.

The invention discloses a satellite/inertial/visual combination navigation system integrity evaluation method, relates to integrity evaluation of a multi-sensor combination navigation system and belongs to the calculating, reckoning and counting technology fields. In the method, a measured value of each sensor at current time is preprocessed to obtain an observation vector of each sensor at the current time, and the observation vector of each sensor at the current time is expressed as multiple edges having a common pose node. A relationship between the pose node and the edges is used, graph optimization is used to construct a residual weighted quadratic sum function, and a firefly algorithm with high robustness is adopted to optimize the pose node to solve an optimal solution of the residual weighted quadratic sum function and realize a global optimum. And then, test statistics is calculated and the test statistics is compared with a detection threshold so as to determine whether there is a fault. A system calculation amount is reduced, a time delay problem caused by sudden increase of memory consumption over time is avoided, and a real-time detection requirement can be satisfied.

A single laser radar synchronous positioning and composition system has the defects of high energy consumption, low sensing resolution and unobvious depth value change. A single vision synchronous positioning and composition system is low in precision and is easily interfered by illumination change. The invention provides an SLAM and target tracking method, which comprises the following steps: (1)collecting three-dimensional point cloud data and image video data in a real urban road scene, and processing the three-dimensional point cloud data and the image video data; (2) performing segmentation according to a processing result to obtain a static scene and a dynamic scene; (3) processing the static scene, outputting a self-positioning and static scene sub-graph, processing the dynamic scene, outputting a dynamic target pose and track, and constructing a dynamic object sub-graph; and (4) finally, outputting the optimized self-positioning and scene land by adopting a graph optimizationglobal optimization method. The method can improve the positioning precision of a driverless vehicle in a large-scale complex scene, build a high-precision scene map with good consistency, and achievethe tracking of a moving target.