CN107818338B - Method and system for building group mode identification facing map synthesis - Google Patents
Method and system for building group mode identification facing map synthesis Download PDFInfo
- Publication number
- CN107818338B CN107818338B CN201710967938.4A CN201710967938A CN107818338B CN 107818338 B CN107818338 B CN 107818338B CN 201710967938 A CN201710967938 A CN 201710967938A CN 107818338 B CN107818338 B CN 107818338B
- Authority
- CN
- China
- Prior art keywords
- building
- graph
- group
- subgraphs
- subgraph
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
- G06V10/267—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion by performing operations on regions, e.g. growing, shrinking or watersheds
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
- G06V10/457—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components by analysing connectivity, e.g. edge linking, connected component analysis or slices
Abstract
The embodiment of the invention discloses a method and a system for identifying building group modes for map synthesis, wherein the method comprises the following steps: dividing the whole building topographic map by using a road network as a global constraint condition to obtain a plurality of building blocks; performing triangle subdivision on each building block in a plurality of building blocks; calculating the topological relation, the length of the skeleton line and the average distance index value of the building based on the constrained triangular network; adopting a depth-first traversal algorithm to the non-connected graph to obtain a plurality of connected subgraphs; performing group pattern recognition on each connected subgraph in the plurality of connected subgraphs based on the trained random forest classifier; and carrying out comprehensive operation processing on the group mode. By implementing the embodiment of the invention, a top-down graph segmentation method is adopted, so that various types of building potential group modes can be obtained, and then a classifier is adopted to judge the potential group modes, so that a large amount of manual parameter setting is avoided.
Description
Technical Field
The invention relates to the technical field of map making synthesis, in particular to a method and a system for identifying building group modes for map synthesis.
Background
Map synthesis is a process of producing a map with a smaller scale by solving a spatial conflict generated in a map scale expansion process through certain comprehensive operations such as selection, shifting, simplification, combination and the like. Map synthesis plays a very important role in understanding and modeling geospatial space, and is an indispensable technology for updating multi-scale spatial database. The building is used as the most important geographic element of a city, and the building synthesis plays an extremely important role in the aspects of multi-scale database construction, user navigation, city planning analysis and the like.
Building synthesis generally comprises two processes, namely building group pattern recognition and synthesis operation execution, and the present invention focuses on the first aspect. A building group pattern refers to an arrangement or a form formed by a group of buildings over a scale space that can be visually perceived and may be named. The types of the patterns are classified into regular and irregular group patterns, wherein the regular group patterns comprise a linear pattern, a rectangular pattern, a grid pattern and the like, and the irregular pattern comprises an L-shaped pattern, a T-shaped pattern, a Z-shaped pattern, a high-density area and the like. The building group pattern recognition is to divide the buildings in one area into combinations of different forms by adopting a certain recognition technology according to the spatial relationship among the buildings. Due to the lack of available formalized multi-type group pattern recognition knowledge, existing methods are limited to recognizing single-type patterns, such as straight-line types or grid types. These methods have limited effect on map synthesis that requires the ability to identify multiple types of group patterns at once. Therefore, there is a need to construct a unified method that can recognize multiple types of building patterns.
Although building crowd identification is critical to map synthesis, accurately identifying various types of building patterns remains a significant challenge, as patterns are scale-dependent and there is a great variance in the distribution of buildings in different areas.
The prior art relates to a method for defining patterns to be identified, such as a straight line pattern, a curve pattern and then performing comparison identification by adopting a template matching technology.
The type of identification method related in the prior art adopts a topology analysis technology to divide groups, namely, each building generates a buffer zone with a certain distance, and the buildings with crossed buffer zones are divided into a combination.
In the prior art, a method for identifying the buildings of one class is also involved, similarity of each pair of adjacent buildings is calculated according to the distance, shape, direction and other relations between the adjacent buildings, then the combination with the highest similarity is used as a combination, and then the combination is performed again according to the maximum similarity of the adjacent buildings between the combinations, and the combination is stopped when the similarity between the adjacent buildings between all the adjacent combinations reaches a set threshold value in a circulating manner, so that the division of the building group is completed.
The building group pattern recognition method is very suitable for map synthesis if it can recognize groups of arbitrary shapes (i.e. multiple types) in an unsupervised manner.
However, the above methods still have some disadvantages, including: due to the lack of available formalized multi-type group pattern recognition knowledge, the above-mentioned methods can only recognize single-type building group patterns, such as straight line patterns, grid patterns, etc., and these methods are developed separately and are difficult to integrate into one system. In addition, in the prior art, more manual experience threshold setting is involved, the method is poor in repeatability, and the universality needs to be further improved. In the prior art, a large amount of similarity calculation of the distance, the direction, the shape and the size between adjacent buildings is still required, and the proximity between the adjacent buildings is difficult to be accurately expressed by using one or more integration indexes.
Disclosure of Invention
The invention aims to overcome the defects of the prior art, and provides a method and a system for identifying a building group mode facing map synthesis.
In order to solve the above problems, the present invention provides a method for identifying a building group pattern for map synthesis, comprising the following steps:
dividing the whole building topographic map by using a road network as a global constraint condition to obtain a plurality of building blocks;
carrying out triangle subdivision on each building block in a plurality of building blocks to generate a constrained triangle network;
calculating the topological relation, the length of the skeleton line and the average distance index value of the building based on the constrained triangular network;
constructing a non-connected graph based on the topological relation of the buildings;
adopting a depth-first traversal algorithm to the non-connected graph to obtain a plurality of connected subgraphs;
performing group mode recognition on each connected subgraph in the plurality of connected subgraphs based on the trained classifier, and clustering the connected subgraphs with group mode attributes into a group mode set;
and after each building block in the plurality of building blocks is judged to be processed, performing comprehensive operation processing on the group mode set by adopting a deleting, merging and simplifying mode.
The triangulating each building block in the plurality of building blocks to generate the constrained triangular net comprises:
performing fixed interval interpolation on roads in each building block and the sides of the buildings, and generating a constrained triangular net by using all line nodes;
and deleting triangles of the connecting road and triangles inside the building to obtain the trimmed constrained triangular net.
In the calculation of the topological relation of the building, the lengths of the skeleton lines and the average distance index value based on the constrained triangular network:
building topological relation: a topological relationship for indicating the presence of two objects based on the proximity relationship indicator, the formula being: r ═ Ri,jWherein: i 1: n, j 1: n, n representing the number of buildings, R i,j0 denotes that the building i is separated from j, R i,j1 denotes that building i is adjacent to j, R i,j2 means that building i is connected to j;
adjacent subject skeleton lines: the skeleton line between adjacent objects is formed by a midpoint connecting line connecting two sides of a triangle between the two objects, and the formula is as follows: l ═ Li，j＝∑li,j,kWherein: li,j,kA midpoint connecting line connecting two sides of a triangle between two adjacent objects is shown, and k represents the kth triangle;
average distance: the average distance between adjacent objects is calculated according to the skeleton line between the adjacent objects, and the formula is as follows:
wherein: di,jRepresents the average distance, h, of neighboring objects i and ji,j,kIndicating the height of the kth triangle between adjacent objects.
The constructing of the non-connected graph based on the building topological relation comprises the following steps:
each building in the building block is taken as a node of the graph, edges among the nodes are expressed by using a space topological relation among the buildings represented by the nodes, the calculation is carried out according to an adjacent relation index in the topological relation of the buildings, if two buildings with adjacent or adjacent relation exist, an edge exists between the corresponding nodes, otherwise, no edge exists; and when the adjacent building distance is greater than the index threshold value, determining that no adjacent relation exists.
The step of performing a depth-first traversal algorithm on the non-connected graph to obtain a plurality of connected subgraphs comprises:
starting from any node in the non-connected graph, storing nodes with edge connection and node edges into a combination, and marking the nodes as traversed so as to obtain a connected subgraph;
then starting from any node which is not necessarily marked, searching all nodes connected with edges, and storing the nodes into another combination so as to obtain another connected subgraph;
and circulating the process to obtain all connected subgraphs in the non-connected subgraphs.
The training-based classifier group pattern recognition for each connected subgraph in the plurality of connected subgraphs comprises:
training with artificially collected samples by adopting a random forest algorithm;
performing group pattern recognition on each connected subgraph in a plurality of connected subgraphs in a trained random forest classifier, wherein the sample attribute is designed according to the continuity, proximity and integrity principles of a form tower organizational law, and the method comprises the following steps: group average distance, average distance standard deviation, black-white ratio, average visual field area, contour coefficient.
The training-based random forest classifier performing group pattern recognition on each connected subgraph in a plurality of connected subgraphs comprises:
reading a first connected subgraph in the plurality of connected subgraphs;
calculating each index feature of the first communication sub-graph;
constructing a classifier prediction instance object according to each index feature of the first connection subgraph;
the input classifier judges, if the judgment meets the group mode attribute, the first connection subgraph is added into the group mode set, and the first connection subgraph is removed from the multiple connection subgraphs;
and reading a second one of the multiple connected graph subgraphs, and continuing group pattern recognition until all the connected graph subgraphs in the multiple connected graph subgraphs complete group pattern recognition.
After the input classifier judges, the method further comprises the following steps:
if the connected first connected subgraph does not meet the group mode attribute, carrying out segmentation processing on the first connected subgraph and obtaining a non-connected graph after the segmentation processing;
and adopting a depth-first traversal algorithm to the non-connected graph after the segmentation processing to obtain a plurality of connected subgraphs, and continuing the group pattern recognition process.
The step of performing segmentation processing on the first connected graph and obtaining a non-connected graph after the segmentation processing comprises:
representing the weight of the edge by using the proportion of the number of the remaining triangles in the constrained triangulation network between the adjacent buildings and the number of the original triangles generated by only two buildings;
in the cutting process of the graph, if the side length of one triangle is larger than the longest side of all the triangles in the graph minus a step value, the triangle is deleted, and the number of the remaining triangles is obtained; if the proportion of the number of the residual triangles to the number of the original triangles is smaller than a set threshold value, namely the weight of the edge is smaller than the set threshold value, the edge is deleted;
and changing into a non-connected graph in a plurality of iterative cutting processes.
Correspondingly, the invention also provides a computer system, which comprises a processor and a control unit, wherein the processor is suitable for realizing each instruction; and a storage device adapted to store a plurality of instructions adapted to be loaded by the processor and to perform the method described above.
In the embodiment of the invention, a top-down graph segmentation method is adopted firstly, so that potential group modes of various types of buildings can be obtained, and the defect of a single type identification method is avoided; then, a classifier is adopted to judge whether the potential group mode is the group mode, so that a large amount of manual parameter setting is avoided; in the graph segmentation process, the method proposes to express the proximity between the adjacent objects by using the proportion of the residual triangles between the adjacent buildings in the constrained triangulation network and the original triangle number generated by only two adjacent objects, so that a large amount of calculation of the relationship between the adjacent objects can be avoided. The invention adopts the random forest classifier, and can identify various types of group modes. The existing methods can only identify one to two modes (such as a straight line type, a curve type and the like) due to setting specific conditions, and the methods for identifying the specific mode are developed in different programs and are difficult to integrate into a system. Accurate and computationally simple proximity expressions are employed. The method expresses the proximity of the adjacent objects by the proportion of the residual triangle quantity and the original triangle quantity, and the index can reflect the spatial relationship such as the distance, the direction and the like between the adjacent objects. The method adopts the automatic graph segmentation and classifier, wherein the related parameters are all automatically acquired without setting empirical parameters.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.
FIG. 1 is a flow chart of a method for map synthesis-oriented building group pattern recognition in an embodiment of the present invention;
FIG. 2 is a schematic diagram of the construction of a building triangulation, a non-connected graph and a graph segmentation process in an embodiment of the invention;
FIG. 3 is a diagram illustrating an example of a decision tree determination process in a classifier according to an embodiment of the present invention.
FIG. 4 is a block diagram of a computer system according to an embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The method for identifying the map synthesis-oriented building group mode can divide a topographic map of the whole building by using a road network as a global constraint condition to obtain a plurality of building blocks; carrying out triangle subdivision on each building block in a plurality of building blocks to generate a constrained triangle network; calculating the topological relation, the length of the skeleton line and the average distance index value of the building based on the constrained triangular network; constructing a non-connected graph based on the topological relation of the buildings; adopting a depth-first traversal algorithm to the non-connected graph to obtain a plurality of connected subgraphs; performing group mode recognition on each connected subgraph in the plurality of connected subgraphs based on a trained random forest classifier, and clustering the connected subgraphs with group mode attributes into a group mode set; and after each building block in the plurality of building blocks is judged to be processed, performing comprehensive operation processing on the group mode set by adopting a deleting, merging and simplifying mode.
The method adopts a top-down graph segmentation method, so that various types of building potential group modes can be obtained, and the defect of a single type identification method is avoided; then, a classifier is adopted to judge whether the potential group mode is the group mode, so that a large amount of manual parameter setting is avoided; in the graph segmentation process, the method proposes to express the proximity between the adjacent objects by using the proportion of the residual triangles between the adjacent buildings in the constrained triangulation network and the original triangle number generated by only two adjacent objects, so that a large amount of calculation of the relationship between the adjacent objects can be avoided.
The embodiment of the invention can complete the building group pattern recognition of a regional topographic map by the following steps, firstly, a preprocessing stage is carried out, namely, roads are used for dividing the topographic map of buildings in the whole region into different blocks, then, a constraint triangulation network is generated for the buildings in each block, the constraint triangulation network is used for calculating the proximity relation, the average distance and the length of a skeleton line between adjacent buildings, and a non-connected graph is constructed by using the buildings and the proximity relation. And then carrying out building group pattern recognition, namely judging a connected subgraph contained in the non-connected graph of the structure by adopting a classifier, judging whether the connected subgraph is a pattern needing to be recognized, if so, taking the connected subgraph as a group pattern, if not, segmenting the connected subgraph, generating a plurality of sub-connected graphs, then judging the sub-connected graphs, and circulating the process until no connected subgraph enters the classifier, and finishing grouping. And finally, synthesizing the building group by adopting different map comprehensive operations according to the building group mode characteristics.
Specifically, fig. 1 shows a flowchart of a method for map synthesis-oriented building group pattern recognition in an embodiment of the present invention, which is suitable for building topographic pattern recognition, and specifically includes the following steps:
s101, dividing a topographic map of the whole building by using a road network as a global constraint condition to obtain a plurality of building blocks;
in a specific implementation process, in order to improve the pattern recognition efficiency, the entire building topographic map may be subdivided using a road network as a global constraint condition, that is, the topographic map may be divided into different regions to obtain a plurality of building blocks (a block is composed of roads and buildings surrounded by the roads). Each block is used as a single subsequent processing unit, i.e. to implement triangulation, i.e. to complete the method process between steps S102 to S109 as a unit.
S102, performing triangle subdivision on each building block in a plurality of building blocks to generate a constrained triangle network;
in a specific implementation process, triangle subdivision is performed on each building block in a plurality of building blocks, in order to accurately calculate the building proximity relation, the skeleton line length and the average distance index value, first, fixed-interval (such as 2m) interpolation is performed on the roads and the building edges in one block, and a constrained triangulation network (as shown in fig. 2a) is generated by using all line nodes, namely, the triangle subdivision. Since not all triangles can be used to calculate the proximity relation and the related index value, such as the triangle connecting the roads and the triangle inside the building, they can be deleted to obtain the pruned constrained triangulation (fig. 2 d).
S103, calculating a topological relation, a skeleton line length and an average distance index value of the building based on the constrained triangular network;
and calculating the index values of the topological relation, the skeleton line length and the average distance of the building by using the pruned constraint triangulation network, and storing the index values into n x n (n represents the number of buildings), wherein the index values are used for the construction of the non-connected graph in the step S104 and the calculation of the new sample characteristics in the step S106.
Proximity relationship: the proximity relation index is used for indicating the topological relation existing in two objects, including adjacent, connected and separated, and is respectively represented by 1, 2 and 0. If a triangle exists between two objects, it is said that their relationship is adjacent; if they are common, the description is connected; otherwise, the relationship is a phase-separated relationship, and the proximity relationship is expressed by the following formula:
R＝Ri,jformula (1)
In the formula: i is 1: n, j is 1: n, and n represents the number of buildings. R i,j0 denotes that the building i is separated from j, Ri,jAnd 1 represents that a building is adjacent to a building, and represents that the buildings i and j are connected.
Adjacent subject skeleton lines: the skeleton line between adjacent objects is formed by a midpoint connecting line connecting two sides of a triangle between the two objects, and is calculated by the following formula:
L＝Li，j＝∑li,j,kformula (2)
In the formula: li,j,kA line connecting midpoints between two sides of a triangle between two adjacent objects is shown, and k represents the kth triangle.
Average distance index value: the average distance between adjacent objects is calculated according to the skeleton line between the adjacent objects, and is calculated by the following formula:
in the formula: di,jRepresents the average distance, h, of neighboring objects i and ji,j,kIndicating the height of the kth triangle between adjacent objects.
S104, constructing a non-connected graph based on the topological relation of the building;
this step abstracts the buildings and their proximity to the structure of the graph, thus converting the processing of the buildings to the processing of the graph. In the construction of the graph, each building in the neighborhood is considered as a node of the graph, and the edges between the nodes are expressed by the spatial topological relationship between the buildings represented by the nodes. According to the proximity relation calculated in S103, if two buildings with adjacency or proximity relation exist, an edge exists between the corresponding nodes, otherwise no edge exists. Some nearby buildings may be considered to be out of proximity because they are a large distance apart (e.g., greater than 20 meters). The graph thus constructed is a non-connected graph (e.g. fig. 2e) comprising a plurality of connected subgraphs (any two nodes can be reached through the relationship of edges), each connected subgraph representing a potential building group pattern.
S105, obtaining a plurality of connected subgraphs by adopting a depth-first traversal algorithm for the unconnected subgraphs;
according to the non-connected graph constructed in S104, a plurality of connected subgraphs (abstract expressions of a plurality of building group sets) can be obtained by using a depth-first traversal algorithm, such as the subgraph1 and the subgraph2 in fig. 2 e. When traversing the non-connected homograph, the algorithm starts from any node, stores nodes and edges with edge connection into a combination, and marks the nodes as traversed, so that a connected subgraph is obtained. Then, starting from any node which is not necessarily marked, all nodes connected with edges are searched, and all nodes are stored in another combination, namely another connected subgraph is obtained. And circulating the process to obtain all connected subgraphs in the non-connected subgraphs.
S106, carrying out group mode recognition on each connected subgraph in the multiple connected subgraphs based on the trained classifier;
in a specific implementation process, reading a first connected subgraph in a plurality of connected subgraphs; calculating each characteristic of the first communication subgraph; constructing a classifier prediction instance object according to each feature of the first connection subgraph; the input classifier judges, if the judgment meets the group mode attribute, the first connection subgraph is added into the group mode set, and the first connection subgraph is removed from the multiple connection subgraphs; and reading a second one of the multiple connected graph subgraphs, and continuing group pattern recognition until all the connected graph subgraphs in the multiple connected graph subgraphs complete group pattern recognition.
S107, judging whether the connected subgraph meets the group mode attribute, if so, entering S109, and if not, entering S108;
the trained random forest classifier is mainly used for judging whether a group pattern needing to be identified exists in the potential building groups represented by the connected subgraph, and two types of judgment results are obtained based on the group pattern identification, namely the existence of the building groups and the nonexistence of the building groups. If the judgment result is yes, adding the connected subgraph into the group list, removing the connected subgraph from the connected subgraph set, and reading the next connected subgraph for judgment; if the judgment result is no, the connected subgraph is divided (namely S108). In the judgment process, a connected subgraph is read firstly, then each feature of the connected subgraph (namely the feature of a group represented by the connected subgraph) is calculated, a classifier prediction instance object is constructed according to the features, and finally the classifier prediction instance object is input into the classifier for judgment. This classifier is trained using a random forest algorithm with manually collected samples. The random forest algorithm is to establish a forest in a random manner, wherein the forest is composed of a plurality of decision trees (for example, 200), and each decision tree has no relation with each other. After the forest is obtained, when a new instance object enters, each decision tree in the forest is judged, and then the instance object is determined to belong to the class according to the class with the maximum judgment results of all the decision trees. Each decision tree decision process is as shown in fig. 3 (an example of a decision tree, where the value is an assumed value), that is, a round of classification may be performed first from the most important attribute according to the attribute of the sample, and when the attribute is determined to be "yes", the next secondary attribute determination is performed, and so on until the final determination result is obtained. And if the judgment result is no, ending the judgment and dividing the group corresponding to the sample. Samples include two classes, positive samples (group mode) and non-samples (non-group mode). The attributes of the samples are designed according to the continuity, proximity and integrity principles of the organizational law of the form tower, and specifically include the following 5:
group average distance: the index is used to reflect the proximity of the group, and the average distance of the group is easily calculated according to formula (3), which is as follows:
in the formula: li,jRepresenting the length of the skeleton line between adjacent buildings i and j, di,jRepresenting the average distance between adjacent buildings i and j.
Mean distance standard deviation: the index is mainly used for indicating the homogeneity of the group, and if the value is larger, the index can be divided again. Calculated according to formula (4), the specific formula is as follows:
in the formula: sdStandard deviation representing group mean distance, n represents number of neighboring pairs of group buildings, diRepresenting the average distance between the ith pair of adjacent buildings.
Black-white ratio: the index is used for indicating the compactness of the group and is expressed by dividing the area of all buildings in the group by the area of the convex hull of the outline of the group, and the calculation formula is as follows:
in the formula: a. theiRepresents the ith building area, ACHRepresenting the area of the convex hull of the group contour.
Average visual field area: the index is used to indicate the compactness of the group and is expressed by dividing the area of the visual field (formed by the constraint triangulation network) among all the adjacent objects by the total area of the building, and the calculation formula is as follows:
in the formula: a. thetiDenotes the ith triangle area, AjThe jth building area.
Contour coefficient: because the building shape has the feature of squaring, the group formed by the building shape also has a certain rectangular characteristic. The index is mainly used for reflecting the integrity of the group, the more the group contour tends to be rectangular, the higher the integrity is, namely the contour coefficient is larger, and the calculation formula is as follows:
in the formula: a. theiRepresents the ith building area, AmbraRepresenting the area of the smallest bounding rectangle of the group.
It should be noted that the above group average distance, average distance standard deviation, black-white ratio, average visible area, and threshold of the contour coefficient in each decision tree of the random forest classifier are determined according to the training samples, and the characteristic threshold and the importance (ranking in the decision tree) of each decision tree may be different, but the determination and identification process is similar to that of fig. 3, that is, the group average distance determination is performed first, then the distance standard deviation determination is performed, then the black-white ratio determination is performed, then the judgment of the area of the parallel visible area is performed, and then the process of the contour coefficient is performed. The set value for the group average distance in fig. 3 is 10m, and of course, within a certain allowable range, different thresholds exist for different group modes, which is not limited to 10m, and for example, any value within a range of 1m to 20m is possible; the values and the sequence of the average distance standard deviation, the black-white ratio, the average visible area, the contour coefficient, etc. are not limited to those shown in fig. 3, and these threshold conditions may be determined according to different group mode attributes.
S108, carrying out segmentation processing on the connected subgraphs which do not meet the group mode attribute to obtain non-connected subgraphs after the segmentation processing;
if a connected subgraph passes through the classifier, the judgment result is negative, that is, the group mode does not exist in the connected subgraph, and the connected subgraph needs to be segmented at this time. The graph is divided by deleting the edges of the graph, and the deletion of the edges is determined according to the weight of the edges. The present embodiment represents the weight of an edge using the ratio of the number of remaining triangles of the neighborhood pair building in the constrained triangulation network to the number of original triangles generated by only two buildings (as in the bottom right diagram of fig. 2 d). During the cutting process of the graph, if the side length of a triangle is larger than the longest side of all the triangles in the graph minus a step value, the triangle is deleted, and the number of the remaining triangles is obtained. If the ratio of the number of remaining triangles to the number of original triangles is less than a set threshold (obtained by sample statistics), i.e. the weight of an edge is less than the set threshold, the edge will be deleted. And (6) changing the connected subgraph into a non-connected subgraph in the process of multiple iterative cutting, then passing the non-connected subgraph through S105 to obtain a plurality of connected subgraphs, and circulating S106 until no connected subgraph enters the classifier to obtain all groups.
S109, grouping the connected subgraphs with the group mode attributes into a group mode set;
s110, judging whether unprocessed connected subgraphs exist, if yes, entering S106 to continue the group pattern recognition process of the connected subgraphs in the plurality of connected subgraphs, and if all the connected subgraphs in the plurality of connected subgraphs are judged to be processed, entering S111;
s111, judging whether an unprocessed building block exists, if so, entering S102 to continue triangulation on building blocks in a plurality of building blocks to generate a constrained triangular net, and if so, entering S112;
s112, forming a final group mode set;
and S113, carrying out comprehensive operation processing on the group mode set by adopting a deleting, merging and simplifying mode.
In the specific implementation process, according to the final group mode set obtained in S112, the deletion, combination, and simplification comprehensive operations are adopted to synthesize the group mode set. Deletion means deleting groups, merging means merging all buildings in a group into one object, and simplification means deleting some redundant details under the condition of saving the overall characteristics of the graph. The selection of the comprehensive operation principle comprises the following steps: 1) deleting the group if the area of the group is smaller than a set threshold; 2) if the number of the groups is more than 1 and the area of the group building is more than a set threshold value, executing merging and simplifying operation; 3) if the number of the groups is 1, the simplification operation is executed.
Correspondingly, fig. 2 also shows the process of dividing the street triangles, constructing the non-connected graph and dividing the graph. Firstly, interpolating the edges and roads of the building, and generating a constraint triangular network (figure 2a) by using all nodes; since not all triangles are useful for calculation, triangles connecting roads and three vertices on a building are deleted, so that a modified constrained triangulation network is obtained (fig. 2e, fig. 1S102), and proximity relation detection, average distance and skeleton line length calculation are performed by using the triangles (fig. 1S 103); constructing a non-connected graph according to the proximity relation (fig. 2f, fig. 1S104), wherein the non-connected graph depth-first traversal algorithm can find connected subgraphs (such as the subgraph1 and the subgraph2 in fig. 2 f) in the non-connected graph; then, each connected subgraph is judged by using a classifier (fig. 1S106 and fig. 1S107), and five characteristics of the group mode corresponding to the connected subgraph are calculated in the judgment process. If the judgment result is yes, adding the sub-images into the group mode set, otherwise, dividing the sub-images and the building groups corresponding to the sub-images (fig. 2c, fig. 1S108), obtaining a plurality of connected sub-images (fig. 2d), and judging the connected sub-images by using the classifier again.
Correspondingly, fig. 4 also shows a computer system configuration diagram, which includes a processor adapted to implement instructions; and a storage device adapted to store a plurality of instructions adapted to be loaded by the processor and to perform the method as specifically referred to:
dividing the whole building topographic map by using a road network as a global constraint condition to obtain a plurality of building blocks;
carrying out triangle subdivision on each building block in a plurality of building blocks to generate a constrained triangle network;
calculating the topological relation, the length of the skeleton line and the average distance index value of the building based on the constrained triangular network;
constructing a non-connected graph based on the topological relation of the buildings;
adopting a depth-first traversal algorithm to the non-connected graph to obtain a plurality of connected subgraphs;
performing group mode recognition on each connected subgraph in the plurality of connected subgraphs based on a trained random forest classifier, and clustering the connected subgraphs with group mode attributes into a group mode set;
and after each building block in the plurality of building blocks is judged to be processed, performing comprehensive operation processing on the group mode set by adopting a deleting, merging and simplifying mode.
Specifically, the triangulating each building block of the plurality of building blocks to generate the constrained triangulated mesh includes: performing fixed interval interpolation on roads in each building block and the sides of the buildings, and generating a constrained triangular net by using all line nodes; and deleting triangles of the connecting road and triangles inside the building to obtain the trimmed constrained triangular net.
Specifically, in the calculation of the building topological relation, the skeleton line length and the average distance index value based on the constrained triangular network:
building topological relation: a topological relationship for indicating the presence of two objects based on the proximity relationship indicator, the formula being: r ═ Ri,jWherein: i 1: n, j 1: n, n representing the number of buildings, R i,j0 denotes that the building i is separated from j, R i,j1 denotes that building i is adjacent to j, R i,j2 means that building i is connected to j;
adjacent subject skeleton lines: the skeleton line between adjacent objects is formed by a midpoint connecting line connecting two sides of a triangle between the two objects, and the formula is as follows: l ═ Li，j＝∑li,j,kWherein: li,j,kA midpoint connecting line connecting two sides of a triangle between two adjacent objects is shown, and k represents the kth triangle;
average distance: the average distance between adjacent objects is calculated according to the skeleton line between the adjacent objects, and the formula is as follows:
wherein: di,jRepresents the average distance, h, of neighboring objects i and ji,j,kIndicating the height of the kth triangle between adjacent objects.
Specifically, the constructing of the non-connected graph based on the building topological relation comprises the following steps: each building in the building block is taken as a node of the graph, edges among the nodes are expressed by using a space topological relation among the buildings represented by the nodes, the calculation is carried out according to an adjacent relation index in the topological relation of the buildings, if two buildings with adjacent or adjacent relation exist, an edge exists between the corresponding nodes, otherwise, no edge exists; and when the adjacent building distance is greater than the index threshold value, determining that no adjacent relation exists.
Specifically, the performing a depth-first traversal algorithm on the non-connected graph to obtain a plurality of connected subgraphs includes: starting from any node in the non-connected graph, storing nodes with edge connection and node edges into a combination, and marking the nodes as traversed so as to obtain a connected subgraph; then starting from any node which is not necessarily marked, searching all nodes connected with edges, and storing the nodes into another combination so as to obtain another connected subgraph; and circulating the process to obtain all connected subgraphs in the non-connected subgraphs.
Specifically, the performing group pattern recognition on each connected subgraph in the plurality of connected subgraphs by using the training-based random forest classifier comprises: training with artificially collected samples by adopting a random forest algorithm; the method comprises the following steps of carrying out group pattern recognition on each connected subgraph in a plurality of connected subgraphs based on a trained random forest classifier, wherein sample attributes are designed according to the continuity, proximity and integrity principles of a form tower organizational law, and the method comprises the following steps: group average distance, average distance standard deviation, black-white ratio, average visual field area, contour coefficient.
Specifically, the performing group pattern recognition on each connected subgraph in the plurality of connected subgraphs by using the training-based random forest classifier comprises: reading a first connected subgraph in the plurality of connected subgraphs; calculating each characteristic of the first communication subgraph; constructing a classifier prediction instance object according to each feature of the first connection subgraph; the input classifier judges, if the judgment meets the group mode attribute, the first connection subgraph is added into the group mode set, and the first connection subgraph is removed from the multiple connection subgraphs; and reading a second one of the multiple connected graph subgraphs, and continuing group pattern recognition until all the connected graph subgraphs in the multiple connected graph subgraphs complete group pattern recognition.
Specifically, the input classifier further includes, after performing the determination: if the connected first connected subgraph does not meet the group mode attribute, carrying out segmentation processing on the first connected subgraph and obtaining a non-connected graph after the segmentation processing; and adopting a depth-first traversal algorithm to the non-connected graph after the segmentation processing to obtain a plurality of connected subgraphs, and continuing the group pattern recognition process.
Specifically, the dividing the first connected graph and obtaining the non-connected graph after the dividing includes: representing the weight of the edge by using the proportion of the number of the remaining triangles in the constrained triangulation network between the adjacent buildings and the number of the original triangles generated by only two buildings; in the cutting process of the graph, if the side length of one triangle is larger than the longest side of all the triangles in the graph minus a step value, the triangle is deleted, and the number of the remaining triangles is obtained; if the proportion of the number of the residual triangles to the number of the original triangles is smaller than a set threshold value, namely the weight of the edge is smaller than the set threshold value, the edge is deleted; and changing into a non-connected graph in a plurality of iterative cutting processes.
In summary, a top-down graph segmentation method is adopted, so that various types of building potential group patterns can be obtained, and the defect of a single type identification method is avoided; then, a classifier is adopted to judge whether the potential group mode is the group mode, so that a large amount of manual parameter setting is avoided; in the graph segmentation process, the method proposes to express the proximity between the adjacent objects by using the proportion of the residual triangles between the adjacent buildings in the constrained triangulation network and the original triangle number generated by only two adjacent objects, so that a large amount of calculation of the relationship between the adjacent objects can be avoided. The invention adopts the random forest classifier, and can identify various types of group modes. The existing methods can only identify one to two modes (such as a straight line type, a curve type and the like) due to setting specific conditions, and the methods for identifying the specific mode are developed in different programs and are difficult to integrate into a system. Accurate and computationally simple proximity expressions are employed. The method expresses the proximity of the adjacent objects by the proportion of the residual triangle quantity and the original triangle quantity, and the index can reflect the spatial relationship such as the distance, the direction and the like between the adjacent objects. The method adopts the automatic graph segmentation and classifier, wherein the related parameters are all automatically acquired without setting empirical parameters.
Those skilled in the art will appreciate that all or part of the steps in the methods of the above embodiments may be implemented by associated hardware instructed by a program, which may be stored in a computer-readable storage medium, and the storage medium may include: read Only Memory (ROM), Random Access Memory (RAM), magnetic or optical disks, and the like.
In addition, the method and the system for identifying the building group pattern facing map synthesis provided by the embodiment of the present invention are described in detail above, and a specific example is applied in the present disclosure to explain the principle and the implementation of the present invention, and the description of the above embodiment is only used to help understanding the method and the core idea of the present invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, there may be variations in the specific embodiments and the application scope, and in summary, the content of the present specification should not be construed as a limitation to the present invention.
Claims (8)
1. A map synthesis-oriented building group pattern recognition method is characterized by comprising the following steps:
dividing the whole building topographic map by using a road network as a global constraint condition to obtain a plurality of building blocks;
carrying out triangle subdivision on each building block in a plurality of building blocks to generate a constrained triangle network;
calculating the topological relation, the length of the skeleton line and the average distance index value of the building based on the constrained triangular network;
constructing a non-connected graph based on the topological relation of the buildings;
adopting a depth-first traversal algorithm to the non-connected graph to obtain a plurality of connected subgraphs;
performing group mode recognition on each connected subgraph in the plurality of connected subgraphs based on a trained random forest classifier, and clustering the connected subgraphs with group mode attributes into a group mode set;
after each building block in the plurality of building blocks is judged to be processed, the group mode set is subjected to comprehensive operation processing in a deleting, merging and simplifying mode;
the triangulating each building block in the plurality of building blocks to generate the constrained triangular net comprises:
performing fixed interval interpolation on roads in each building block and the sides of the buildings, and generating a constrained triangular net by using all line nodes;
deleting triangles of the connecting roads and triangles inside the building to obtain a trimmed constrained triangular net;
in the calculation of the topological relation of the building, the lengths of the skeleton lines and the average distance index value based on the constrained triangular network:
building topological relation: a topological relationship for indicating the presence of two objects based on the proximity relationship indicator, the formula being: r ═ Ri,jWherein: i 1: n, j 1: n, n representing the number of buildings, Ri,j0 denotes that the building i is separated from j, Ri,j1 denotes that building i is adjacent to j, Ri,j2 means that building i is connected to j;
adjacent subject skeleton lines: the skeleton line between adjacent objects is formed by a midpoint connecting line connecting two sides of a triangle between the two objects, and the formula is as follows: l ═ Li，j＝∑li,j,kWherein: li,j,kA midpoint connecting line connecting two sides of a triangle between two adjacent objects is shown, and k represents the kth triangle;
average distance: the average distance between adjacent objects is calculated according to the skeleton line between the adjacent objects, and the formula is as follows:
wherein: di,jRepresents the average distance, h, of neighboring objects i and ji,j,kIndicating the height of the kth triangle between adjacent objects.
2. The map-integrated oriented building complex pattern recognition method of claim 1, wherein constructing a non-connected graph based on building topological relations comprises:
each building in the building block is taken as a node of the graph, edges among the nodes are expressed by using a space topological relation among the buildings represented by the nodes, the calculation is carried out according to an adjacent relation index in the topological relation of the buildings, if two buildings with adjacent or adjacent relation exist, an edge exists between the corresponding nodes, otherwise, no edge exists; and when the adjacent building distance is greater than the index threshold value, determining that no adjacent relation exists.
3. The method for map synthesis-oriented building group pattern recognition as claimed in claim 2, wherein the performing a depth-first traversal algorithm on the non-connected graph to obtain a plurality of connected subgraphs comprises:
starting from any node in the non-connected graph, storing nodes with edge connection and node edges into a combination, and marking the nodes as traversed so as to obtain a connected subgraph;
then starting from any node which is not necessarily marked, searching all nodes connected with edges, and storing the nodes into another combination so as to obtain another connected subgraph;
and circulating the process to obtain all connected subgraphs in the non-connected subgraphs.
4. The method of map synthesis-oriented building group pattern recognition of claim 3, wherein the training-based classifier group pattern recognition for each connected subgraph of a plurality of connected subgraphs comprises:
training with artificially collected samples by adopting a random forest algorithm;
the method comprises the following steps of carrying out group pattern recognition on each connected subgraph in a plurality of connected subgraphs based on a trained random forest classifier, wherein sample attributes are designed according to the continuity, proximity and integrity principles of a form tower organizational law, and the method comprises the following steps: group average distance, average distance standard deviation, black-white ratio, average visual field area, contour coefficient.
5. The method of map synthesis-oriented building group pattern recognition of claim 4, wherein the training-based random forest classifier group pattern recognition for each connected subgraph of a plurality of connected subgraphs comprises:
reading a first connected subgraph in the plurality of connected subgraphs;
calculating each index feature of the first communication sub-graph;
constructing a classifier prediction instance object according to each index feature of the first connection subgraph;
the input classifier judges, if the judgment meets the group mode attribute, the first connection subgraph is added into the group mode set, and the first connection subgraph is removed from the multiple connection subgraphs;
and reading a second one of the multiple connected graph subgraphs, and continuing group pattern recognition until all the connected graph subgraphs in the multiple connected graph subgraphs complete group pattern recognition.
6. The map-integrated oriented building group pattern recognition method of claim 5, wherein the inputting the classifier after the determining further comprises:
if the connected first connected subgraph does not meet the group mode attribute, carrying out segmentation processing on the first connected subgraph and obtaining a non-connected graph after the segmentation processing;
and adopting a depth-first traversal algorithm to the non-connected graph after the segmentation processing to obtain a plurality of connected subgraphs, and continuing the group pattern recognition process.
7. The method for map synthesis-oriented building group pattern recognition according to claim 6, wherein the step of performing segmentation processing on the first connected sub-graph and obtaining a non-connected graph after the segmentation processing comprises:
representing the weight of the edge by using the proportion of the number of the remaining triangles in the constrained triangulation network between the adjacent buildings and the number of the original triangles generated by only two buildings;
in the cutting process of the graph, if the side length of one triangle is larger than the longest side of all the triangles in the graph minus a step value, the triangle is deleted, and the number of the remaining triangles is obtained; if the proportion of the number of the residual triangles to the number of the original triangles is smaller than a set threshold value, namely the weight of the edge is smaller than the set threshold value, the edge is deleted;
and changing into a non-connected graph in a plurality of iterative cutting processes.
8. A computer system comprising a processor adapted to implement instructions; and a storage device adapted to store a plurality of instructions adapted to be loaded by the processor and to perform the method of any of claims 1-7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710967938.4A CN107818338B (en) | 2017-10-16 | 2017-10-16 | Method and system for building group mode identification facing map synthesis |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710967938.4A CN107818338B (en) | 2017-10-16 | 2017-10-16 | Method and system for building group mode identification facing map synthesis |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107818338A CN107818338A (en) | 2018-03-20 |
| CN107818338B true CN107818338B (en) | 2021-04-06 |
Family
ID=61608005
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710967938.4A Active CN107818338B (en) | 2017-10-16 | 2017-10-16 | Method and system for building group mode identification facing map synthesis |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107818338B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111598036B (en) * | 2020-05-22 | 2021-01-01 | 广州地理研究所 | Urban group geographic environment knowledge base construction method and system of distributed architecture |
| CN111881964A (en) * | 2020-07-20 | 2020-11-03 | 南宁师范大学 | Linear building mode identification method and system based on Delaunay triangulation network |
| CN113190639B (en) * | 2021-05-13 | 2022-12-13 | 重庆市勘测院 | Comprehensive drawing method for residential area |
| CN115578538B (en) * | 2022-10-17 | 2023-06-23 | 北京世冠金洋科技发展有限公司 | Three-dimensional scene generation method and device, electronic equipment and storage medium |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101510311A (en) * | 2009-03-05 | 2009-08-19 | 浙江大学 | Method for rapidly sorting a large amount of building side elevation images based on GPS information |
| CN102663957A (en) * | 2012-03-08 | 2012-09-12 | 北京师范大学 | Automatic generation method of interactive three dimensional city panoramic map |
| CN104063893A (en) * | 2014-04-14 | 2014-09-24 | 北京师范大学 | Urban building visualization method based on Gestalt psychological criterions and multi-tag graph cut minimization |
| CN105957148A (en) * | 2016-05-20 | 2016-09-21 | 江苏得得空间信息科技有限公司 | Granularity balance data organization method of complicated three-dimensional building model |
| CN106204446A (en) * | 2016-07-01 | 2016-12-07 | 中国测绘科学研究院 | The building of a kind of topography merges method |
-
2017
- 2017-10-16 CN CN201710967938.4A patent/CN107818338B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101510311A (en) * | 2009-03-05 | 2009-08-19 | 浙江大学 | Method for rapidly sorting a large amount of building side elevation images based on GPS information |
| CN102663957A (en) * | 2012-03-08 | 2012-09-12 | 北京师范大学 | Automatic generation method of interactive three dimensional city panoramic map |
| CN104063893A (en) * | 2014-04-14 | 2014-09-24 | 北京师范大学 | Urban building visualization method based on Gestalt psychological criterions and multi-tag graph cut minimization |
| CN105957148A (en) * | 2016-05-20 | 2016-09-21 | 江苏得得空间信息科技有限公司 | Granularity balance data organization method of complicated three-dimensional building model |
| CN106204446A (en) * | 2016-07-01 | 2016-12-07 | 中国测绘科学研究院 | The building of a kind of topography merges method |
Non-Patent Citations (3)
| Title |
|---|
| A Gestalt rules and graph-cut-based simplification framework forurban building models;Yuebin Wang 等;《International Journal of Applied Earth Observation and Geoinformation》;20151231;第247-254页 * |
| Building pattern recognition in topographic data: examples on collinear and curvilinear alignments;Xiang Zhang 等;《Springer》;20111028;正文第1-30页 * |
| SEMANTIC CLASSIFICATION OF URBAN BUILDINGS COMBINING VHR IMAGES AND GIS DATA;S.Du 等;《ISPRS J. Photogramm. Rem. Sens》;20151231;正文第1—4页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107818338A (en) | 2018-03-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107818338B (en) | Method and system for building group mode identification facing map synthesis | |
| Zhang et al. | Building pattern recognition in topographic data: examples on collinear and curvilinear alignments | |
| WO2020233152A1 (en) | Urban building space data-based built-up area boundary identification method and device | |
| CN112070769B (en) | Layered point cloud segmentation method based on DBSCAN | |
| Du et al. | Extracting building patterns with multilevel graph partition and building grouping | |
| CN110992473B (en) | Tree branch modeling method and system based on vehicle-mounted laser scanning point cloud | |
| CN113724279B (en) | System, method, equipment and storage medium for automatically dividing traffic cells into road networks | |
| CN110008602B (en) | Road network selection method considering multi-feature coordination under large scale | |
| WO2019019653A1 (en) | Device and method for extracting topographical boundary | |
| CN107330734A (en) | Business address system of selection based on Co location patterns and body | |
| CN113033516A (en) | Object identification statistical method and device, electronic equipment and storage medium | |
| CN115713605A (en) | Commercial building group automatic modeling method based on image learning | |
| CN113723715A (en) | Method, system, equipment and storage medium for automatically matching public transport network with road network | |
| CN111027574A (en) | Building mode identification method based on graph convolution | |
| CN114693855B (en) | Point cloud data processing method and device | |
| CN116071722A (en) | Lane geometric information extraction method, system, equipment and medium based on road section track | |
| CN116305436A (en) | Existing bridge monitoring method based on combination of three-dimensional laser scanning and BIM | |
| CN114003623A (en) | Similar path typhoon retrieval method | |
| CN116363319B (en) | Modeling method, modeling device, equipment and medium for building roof | |
| CN111881964A (en) | Linear building mode identification method and system based on Delaunay triangulation network | |
| CN116662468A (en) | Urban functional area identification method and system based on geographic object space mode characteristics | |
| CN111080080A (en) | Method and system for estimating risk of geological disaster of villages and small towns | |
| CN113569331B (en) | Building three-dimensional model semantization method and system | |
| CN116416377A (en) | Identification method, device and system for machining characteristics of thin-wall tube laser cutting | |
| CN115661398A (en) | Building extraction method, device and equipment for live-action three-dimensional model |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20211111 Address after: 510399 No. 135, Xingang West Road, Guangzhou, Guangdong Patentee after: Sun Yat-sen University Address before: 510275 No. 135, Xingang West Road, Haizhu District, Guangzhou, Guangdong Patentee before: Xin Qinchuan Patentee before: Zhang Xinchang Patentee before: He Xianjin |
|
| TR01 | Transfer of patent right |