65 results about "Skeleton graph" patented technology

The invention provides a method for segmenting a 3D model, comprising a bounded plane generating step, a skeleton graph extracting step and a skeleton graph segmenting step, wherein the bounded plane generating step is used for processing all the triangles included in the 3D model according to the input triangle mesh data of the 3D model to generate at least a bounded plane suitable for segmenting the 3D model; the skeleton graph extracting step is used for extracting the skeleton graph of the 3D model by the generated bounded plane; and the skeleton graph segmenting step is used for segmenting the extracted skeleton graph into a subgraph or at least two subgraphs which do not overlap with each other according to the information of the generated bounded plane and the information of the vertex adjacent graph of the 3D model. The invention also provides a device for segmenting the 3D model and an image processing system with the device. The method, the device and the system of the invention can improve the precision and efficiency of segmentation of the 3D model.

The invention provides a skeleton graph regression-based three-dimensional human body posture estimation method. The method mainly comprises the steps of segmentation, regression and matching. The process of the method comprises the following steps of firstly, giving an RGB image of a person, generating a foreground skeleton graph and a background skeleton graph through the deconvolution process of an encoder-decoder architecture for each configuration of a cutting scale and a width ruler, respectively feeding a skeleton mapping into a separate regression network, adopting the skeleton graphsas the input, outputting an assumption of three-dimensional gestures so as to generate a plurality of three-dimensional assumptions, and finally selecting an assumption minimum in projection error fortwo-dimensional joint detection as a final output in order to match a two-dimensional observation value. According to the invention, a regression network is independently trained based on skeleton graphs. When the regression network is combined with a plurality of assumptions, the better estimation effect can be achieved on indoor and outdoor data sets. The influences on results caused by illumination, shielding and the like is greatly reduced. The performance of the attitude estimation is improved to a great extent.

A method for automatically transferring attributes between computer-generated models. The method includes storing in memory first and second models represented by polygonal meshes and storing a set of attributes for the first model. A processor operates or runs a compressed graph generator to process the first and second models to generate first and second compressed graphs that are compressed versions of the models. The method includes comparing topological connectivity of the first and second compressed graphs. When the connectivity is similar, the method includes transferring at least a portion of the attributes from the first model to the second model. The compressed graphs may be motorcycle graphs, skeleton graphs, or other forms of compressed graphs. The method includes determining a pair of vertices in the first compressed graph that match vertices in the second compressed graph for use as starting locations in comparing topological connectivity of the compressed graphs.

The invention discloses a gait recognition method based on skeleton map sequence abnormal joint repair. The method comprises the following steps: (1) extracting a skeleton map sequence of a gait target to be detected from a video by using an attitude estimation algorithm; (2) carrying out error correction and smoothing processing on the skeleton graph sequence according to the topological structure space constraint and the context time constraint of the human skeleton; (3) taking the X-axis coordinate sequence and the Y-axis coordinate sequence of the processed skeleton map as input of a network model, and sending the network model to a trained CNN network for space-time feature vector extraction; and (4) classifying the extracted feature vectors by using a Softmax classifier, and identifying the gait identity. According to the gait recognition method, the problem that traditional gait recognition is easily affected by object shielding, clothes, visual angles, carried objects and otherco-variable factors can be avoided, and meanwhile the problem that the recognition rate is not high due to a gait recognition method based on posture detection is solved.

The present invention discloses a figure segmentation method based on a depth map. The method comprises the concrete steps: (1), collecting a skeleton graph, a color image and a depth map at the same time through adoption of a Kinect sensor, and finding out a human body bone area; (2) mapping node coordinates of the skeleton graph area into the depth map, finding out a minimum rectangular area including all the skeleton nodes, and determining the initial figure area in the depth map; (3) performing area iteration expansion of the initial figure area based on the depth information, and obtaining a final rectangular area including a complete figure object; and (4) taking the rectangular area including the complete figure object as a seed area, designing an energy function according to the depth information and the color information of the image pixels, and obtaining a final figure object segmentation result through adoption of a modulated GrabCut algorithm. The figure segmentation method based on a depth map is able to obtain a rectangular area including a whole figure object through adoption of depth information to obtain an accurate and complete figure object segmentation result.

The invention provides a user identity recognition method and system combined with user gait information, and the method comprises the steps: carrying out the posture detection of each frame of pedestrian object in a video sequence of an original data set through a two-dimensional posture estimation system, and extracting the posture information; preprocessing the extracted joint coordinate sequence to generate a human skeleton data set; and finally, constructing a space-time diagram convolutional network model, dividing the skeleton diagram into six sub-diagrams, sharing joints among the sixsub-diagrams, learning an identification model by using the diagram convolutional network, training by using the constructed data set, and optimizing network parameters by using a multi-loss strategycombining classification loss and comparison loss and random gradient descent. And predicting the accuracy of the trained model by using the verification set. According to the method, effective information of the joint points is fully utilized, the motion state in the time dimension is reserved as much as possible, high robustness is achieved for clothes changes and carrying states, and good generalization capacity is achieved on a cross-view task.

The invention belongs to the technical fields of pattern recognition and computer vision, and particularly discloses a rapid graph matching and recognition method based on skeleton graphs. The graph recognition accuracy is guaranteed, and meanwhile the graph matching speed is increased. The rapid graph matching and recognition method includes the implementation steps of (1) extracting outlines of graphs through a classic canny edge detection operator; (2) sampling the outlines of the graphs at equal intervals; (3) splitting the graphs with the sampling points as peaks of triangles; (4) building the internal skeleton structure graphs of the graphs; (5) extracting feature vectors of the skeleton graphs of the graphs; (6) calculating a matching cost matrix of the graphs; (7) finding optimal matching points of the graphs; (8) rotating the skeleton structure graphs of the graphs, and repeating the step (6) and the step (7); (9) outputting the minimum matching cost for serving as the similarity distance of the graphs. By means of the rapid graph matching and recognition method, under the condition that the certain recognition accuracy is maintained, the complexity of a shape descriptor operator is effectively lowered, and the graph matching speed is increased.

Provided is a tree model generating method according to an embodiment of the present invention. The tree model generating method includes separating and extracting a skeleton of a tree from an image including the tree to generate a matte image, applying a graph representation scheme of a two-dimensional (2D) object to the matte image to generate an object skeleton graph, generating a depth information value in accordance with a predetermined rule and allocating the generated depth information value to each node on the object skeleton graph to generate a three-dimensional (3D) primary tree model, adding branches to the primary tree model based on a branch library constructed in advance and the predetermined rule to generate a secondary tree model, and performing self-growth representation with respect to the secondary tree model based on a distance between each node on the object skeleton graph to generate a tertiary tree model.

The invention provides a human skeleton action recognition method based on generalized graph convolution and reinforcement learning. According to the method, a human skeleton sequence matrix is constructed, a predefined skeleton diagram is constructed, a training set is sent to a generalized graph convolutional network for feature extraction, features are aggregated by using global average pooling, the features are classified by using a full connection layer classifier, and network parameters are updated according to a loss function; based on the trained generalized graph convolutional network, the classifier and the features learned by generalized graph convolution, constructing a feature selection network to adaptively select features useful for recognition in the time dimension, and performing training by using a reinforcement learning method. According to the method, a generalized graph convolutional network is designed for a human skeleton action recognition task and is used for capturing related dependence between any nodes so as to extract richer associated features between the nodes. Meanwhile, a feature selection network is designed and used for selecting features useful for recognition in the time dimension, and therefore more accurate action recognition is achieved.

The invention discloses a mammal posture recognition method based on a body contour and a leg joint skeleton, and the method comprises the two steps: 1, carrying out the edge extraction of a wild mammal image in a complex field environment through Outline Mask R-CNN, and obtaining an animal peripheral contour; and based on the contour map, constructing a Tiny VGG light convolutional neural networkfor posture classification of the wild mammal. In the second part, for a wild mammal video sequence in a complex field environment, LEAP is used for rapidly tracking the positions of animal leg jointpoints, and a skeleton diagram is formed; the method is characterized in that changes of leg joint included angles in an animal skeleton diagram serve as features, LSTM serves as a classifier, and the method is used for posture classification of wild mammals. Finally, results obtained by the two models are fused, different postures of the wild mammal are further recognized to achieve the purposeof distinguishing simple daily behaviors of animals, and the method has a certain application prospect.

The invention provides a virtual reality roaming system and method based on improved redirection walking, and the method comprises the steps: inputting a virtual space plane graph, and employing a skeleton graph generation algorithm based on a Voronoi graph to calculate a feasible skeleton graph of a user in a virtual space in advance; according to the skeleton map in the virtual space, establishing a mapping relation between the user feasible skeleton map in the virtual space and the user feasible skeleton map in the physical space by using a static skeleton map mapping algorithm; tracking the spatial position and direction of the user in the physical space in real time through the input module; and according to the established mapping relationship between the virtual space and the physical space, calculating the corresponding spatial position and direction of the user in the virtual space in real time by using a position inverse mapping algorithm and a curvature gain redirection technology, and generating a corresponding user view image through a rendering module.

The invention relates to the technical field of human body behavior recognition, in particular to a human body behavior recognition method based on a multi-scale attention graph convolutional network, which comprises the following steps: acquiring a to-be-recognized original 3D skeleton graph sequence; inputting the original 3D skeleton diagram sequence into a pre-trained human body behavior recognition model; firstly, extracting joint information, skeleton information and motion information from the original 3D skeleton diagram sequence through a multi-branch input module to serve as behavior feature data; then, enabling a multi-scale attention graph convolution module to learn correlation of 3D skeleton joint points based on the behavior feature data, and extracting time sequence information of various behaviors in different duration time; and finally, identifying human body behaviors corresponding to the original 3D skeleton graph sequence through a global attention pooling layer; and outputting a corresponding human body behavior recognition result. The human body behavior recognition method can give consideration to the accuracy and efficiency of human body behavior recognition, so that the recognition effect of human body behavior recognition can be ensured.

Provided is a tree model generating method according to an embodiment of the present invention. The tree model generating method includes separating and extracting a skeleton of a tree from an image including the tree to generate a matte image, applying a graph representation scheme of a two-dimensional (2D) object to the matte image to generate an object skeleton graph, generating a depth information value in accordance with a predetermined rule and allocating the generated depth information value to each node on the object skeleton graph to generate a three-dimensional (3D) primary tree model, adding branches to the primary tree model based on a branch library constructed in advance and the predetermined rule to generate a secondary tree model, and performing self-growth representation with respect to the secondary tree model based on a distance between each node on the object skeleton graph to generate a tertiary tree model.

The invention discloses a skeleton action recognition method based on graph convolution. A basic unit of the skeleton action recognition method is a space-time graph convolution module. The space-time diagram convolution module comprises the following steps: acquiring a skeleton video, constructing a skeleton graph based on each frame of skeleton video, defining different human body part combinations according to the skeleton diagrams, constructing a joint point relation graph for each human body part combination, and further constructing a multi-dimensional relation interaction graph which comprises a component combination interaction dimension and a joint point interaction dimension; carrying out graph convolution on the multi-dimensional interaction graph on the joint point interaction dimension and the component combination interaction dimension; and then sending spatial features obtained by convolution of the two graphs to a local convolution network of time slices to obtain time dynamic features. A plurality of space-time graph convolution modules are stacked in a network model to construct a neural network, and a softmax classifier is used for classification.

The invention discloses a skeleton extraction and feature recognition method for a building plane traffic space, wherein the method comprises the steps: firstly drawing the contour of the building plane traffic space through a polyline, and marking an emergency exit; using a scanning line algorithm to calculate and obtain the division areas of the Thiessen polygon corresponding to the polylines; checking the Thiessen polygon one by one, deleting the Thiessen polygon if the line is located outside the plane contour, checking the dashed line of the Thiessen polygon one by one, and deleting the dashed line if one end of the dashed line is located on the plane contour; then, calculating the price of each node on the dotted line, representing the end corridor as a one-price node, and extending the end point until the end point intersects with the plane contour, the dotted line being a skeleton diagram corresponding to the plane contour; and extracting recognition features on the skeleton drawing. According to the method, the skeleton and the features of the building plane traffic space are automatically identified by means of computational graphics, the feature result is accurate and rapid, and mistakes and omissions caused by manual judgment are avoided.

Disclosed is an apparatus for generating 3D tree images, which comprises an object separation unit which is configured to extract a background and objects from an externally received 2D tree image and to separate a separation target object from the objects; a pattern matching unit which is configured to extract skeleton pattern information from the separation target object and to generate a pattern matching information by matching the skeleton pattern information with the tree shape pattern information which is previously stored in an object pattern database (DB); and a 3D simulation unit modeling simulation unit which is configured to generate a 3D virtual tree model with the aid of the pattern matching information and to express the generated virtual tree model by way of a self-growth simulation.

The invention discloses a multi-view human skeleton automatic labeling method based on OpenPose, belongs to the technical field of unmanned driving, and overcomes the defects of long time consumption,high cost, nonstandard labeling and the like due to the fact that most of existing public data sets are manually labeled. According to the invention, the collected multi-view data is labeled, and data reserve is provided for multi-view pedestrian action recognition model training. The method comprises the following steps: firstly, reading acquired multi-view video data, then performing pedestriandetection through an improved Yolov3 network, and filtering out pictures which do not contain pedestrians; cutting and extracting a detected human body bounding box (bbox) to generate a new picture image-c, and displaying the new picture image-c; and sequentially inputting the image-c into an OpenPose human skeleton extraction network, removing the influence of a complex background, complementingand screening different missing conditions of the skeleton diagram by using different methods, and finally outputting a complete skeleton diagram.

The invention relates to an action recognition method based on unsupervised graph sequence predictive coding and a storage medium, the action recognition method comprises model training and use, is used for recognizing various actions performed by a human body in a skeleton sequence, and aims to solve the problems that an existing action recognition method highly depends on a large amount of labeled data. The precision is low under the condition that only a small amount of labels exist, and an existing unsupervised method does not utilize topological information of a graph in overfitting and is poor in serious generalization ability. The method of the system comprises: carrying out view angle invariant transformation, resampling and block-level skeleton graph data enhancement on skeleton sequence data; enabling embedding of space-time diagram convolution skeleton sequence block to express extraction; aggregating context features by the graph convolutional recurrent neural network; constructing positive and negative sample pairs through predictive coding; and extracting features through the pre-training model, and obtaining an action category corresponding to the to-be-recognized skeleton sequence by using the classifier. Compared with the prior art, the method has the advantages of low training difficulty, high recognition precision, excellent performance and the like.

The invention discloses a skeleton action recognition method based on learnable PL-GCN and ECLSTM, and relates to the field of action recognition. The problems that in the skeleton action recognition process, the feature capture capacity of key frames and significant motion joints is limited, and the similar action classification capacity is weak can be solved. The method comprises the steps that a learnable graph convolutional network (PL-GCN) is provided for the problem that similar action recognition is prone to confusion, and the learnable graph convolutional network (PL-GCN) is used for improving the physical structure of a model; for the problem of weak key frame capture capability, a feature enhanced long and short time memory network (ECLSTM) is provided for enhancing time sequence features; building a skeleton graph by utilizing a graph topological structure of the skeleton sequence data; fusing the spatial features from the image after convolution and the time sequence features extracted by the ECLSTM network; and carrying out average pooling and convolution on the fused features, and then carrying out final feature classification. The method provided by the invention is superior to some current methods in action recognition progress, algorithm complexity and feature extraction capability.

A medium and large quadruped animal behavior recognition method based on a framework search graph convolutional network comprises the following steps: firstly, based on animal skeleton behavior feature extraction, aiming at medium and large quadruped animal video images in a complex field environment, using a pose estimation algorithm DeepLabCut to quickly track the positions of animal body part joint points, forming a space-time skeleton graph, and carrying out space-time feature extraction on the basis of the space-time skeleton graph; and the spatial-temporal characteristics of different behaviors of the quadruped animal are captured. Then, a plurality of space-time diagram convolution operation modules based on animal skeletons are designed, a graph-based search space is formed, residual connection, a bottleneck structure and various attention mechanisms are fused, and the network is lighter while the performance of the recognition model is improved. And then, realizing the continuity of a search space based on a microarchitecture search strategy so as to automatically search a low-cost space-time diagram convolution model for behavior identification of medium and large quadruped animals, and finally realizing the purpose of distinguishing daily behaviors of the animals, thereby having a certain application prospect.

The present disclosure relates to a method for the lightweighting and / or designing of an additively manufactured article. The disclosure further relates to a computer program product adapted for executing the method of the present disclosure as well as an additively manufactured article obtainable by the method according to the present disclosure. The method includes the step(s) of infilling and / or building each of the one or more integral article parts with a quasi-crystalline structure. The additively manufactured article obtainable by performing the method includes a quasi-crystalline structure and / or a quasiperiodic minimal surface infill and / or quasiperiodic minimal surface design structure and / or aperiodic minimal surface design structure and / or aperiodic minimal surface infill. The disclosure further relates to a method of use of a skeleton graph for a preprocessing in an additive manufacturing process.

The invention discloses a crowd-based event detection method and a related device. The method comprises the steps: determining a to-be-detected area in a current frame image, and detecting a reference area in each frame of reference image according to the to-be-detected area, wherein the reference image is a preset number of frame images adjacent to the current frame image in time, and the reference area corresponds to the position of the to-be-detected area in the image; extracting a space crowd skeleton graph in the to-be-detected area and each detected reference area; sequentially connecting the mentioned spatial crowd skeleton maps in the to-be-detected areas and all the reference areas according to a time arrangement sequence to obtain space-time crowd skeleton maps corresponding to the to-be-detected areas; and carrying out feature extraction on the space-time crowd skeleton graph, and carrying out event classification based on the extracted features. According to the technical scheme provided by the invention, the influence of environmental influence on the event detection of the crowd can be reduced, so that the accuracy and applicability of event detection of the crowd are improved.

The invention discloses a skeleton graph human body behavior identification method and system based on multi-stream fusion, and the method comprises the steps: extracting four different data streams from video skeleton data, carrying out the network model training of the four different data streams, and obtaining four different training models; according to the human body behavior recognition method, human body skeleton point data is used as input, four different data streams are processed for model training, so that the network is more sensitive to channel information expressing different motions, and the human body behavior recognition efficiency is improved. Motion features are enhanced, and a behavior recognition model is trained through stacking of a multi-layer space-time diagram convolutional network structure; and finally, model results of four-stream training are fused, so that model outputs are mutually reinforced, and therefore, behavior action categories can be predicted more accurately.

The invention discloses a sports video action recognition method based on an action hotspot map. The sports video action recognition method comprises the following steps: obtaining first five frames of images in each second through a Kinect technology; firstly obtaining a skeleton graph of each group of first frame images, then combining a plurality of subsequent frame images in one group to obtain an action hotspot graph of the group, calculating the motion distance of the action hotspot graph, comparing and analyzing the calculation result with a standard library, identifying whether the action of a section of video is standard or not; and marking and storing non-standard action. The sports video action recognition method has the real-time performance of video image processing, and alsohas good operability and an accurate analysis result.

The invention relates to the technical field of image relation recognition, and provides a social relation recognition method and device, electronic equipment and a storage medium. The method includes obtaining a to-be-identified image; extracting first human body key point information of a first target person and second human body key point information of a second target person in the to-be-identified image; according to the first human body key point information and the second human body key point information, building key point graph data and skeleton graph data between the first target person and the second target person respectively; extracting key point interaction features of the key point graph data through a preset first graph convolution model, and extracting skeleton interaction features of the skeleton graph data through a preset second graph convolution model; and fusing and classifying the key point interaction features and the skeleton interaction features to obtain a social relationship between the first target person and the second target person. According to the invention, the social relation identification accuracy can be improved.

The embodiment of the invention discloses a three-dimensional printing method and device, computer equipment and a storage medium. The method comprises the steps of: acquiring a two-dimensional profile diagram of a three-dimensional model; generating a skeleton diagram of the two-dimensional profile diagram, and obtaining a coordinate corresponding relation between the skeleton diagram and the two-dimensional profile diagram; segmenting the skeleton diagram; determining segmented three-dimensional model parts corresponding to various lines in the skeleton diagram according to the coordinate corresponding relation, and determining the lines as direction vectors of the corresponding three-dimensional model parts; determining slice layering angles of the corresponding three-dimensional model parts according to the direction vectors; and printing the three-dimensional model according to the slice layering angles. According to the technical scheme provided by the embodiment of the invention, the requirements of the three-dimensional model, especially the three-dimensional model with a smaller angle with the horizontal direction, for printing support are effectively reduced, so that the use of printing consumables is reduced, the smoothness and flatness of the surface of the three-dimensional model are improved, and the three-dimensional model is more attractive.

The invention discloses a questionnaire data analysis method based on linear hidden variables. The method comprises the following steps: collecting filled questionnaires, and carrying out preprocessing and standardization processing on the questionnaires; constructing a measurement model, and obtaining a cluster of the observation variables and a skeleton diagram of hidden variables according to the observation variables after standardization processing; enumerating equivalence classes of the hidden variable skeleton graph, and carrying out tripartite constraint judgment; if the three-body constraint is violated, refusing to carry out connection; if the three-split constraint is met, taking each hidden variable as a root node, eliminating the influence from the root node on the rest hidden variables, and retaining corresponding equivalence classes; merging the reserved equivalence classes, outputting a causal structure diagram of the hidden variables according to a merging result, and obtaining a causal relationship among the hidden variables in the questionnaire. According to the method, the causal relationship among the hidden variables distributed in any form can be obtained, auxiliary analysis is carried out on the questionnaire, the analysis result is more accurate, and a correct decision can be made.

The invention discloses a bridge crack detection method based on image superposition and crack information fusion. The method comprises the steps of collecting a bridge image, carrying out the Gaussian smoothing, calculating the image gradient through employing a Sobel operator, extracting a symbiotic edge and a seed point, connecting the seed point, deleting an overlong seed point connection line, obtaining a crack skeleton diagram, deleting an overshort skeleton, and obtaining a crack diagram; acquiring crack information of a single image, superposing the images according to a shooting sequence, connecting cracks at the superposed edges of the images, and acquiring width, length and number information of the cracks on a complete component. The crack image is acquired by adopting the image superposition method, so that the method has the characteristics of high detection precision and high detection speed, and can be used for acquiring the position, width and length information of thecracks on the bridge component.