CN115439877A - Human skeleton recognition method and system based on template matching - Google Patents

Human skeleton recognition method and system based on template matching Download PDF

Info

Publication number
CN115439877A
CN115439877A CN202210882051.6A CN202210882051A CN115439877A CN 115439877 A CN115439877 A CN 115439877A CN 202210882051 A CN202210882051 A CN 202210882051A CN 115439877 A CN115439877 A CN 115439877A
Authority
CN
China
Prior art keywords
template
real
time
point
edge
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.)
Pending
Application number
CN202210882051.6A
Other languages
Chinese (zh)
Inventor
肖德贵
李佳
李健芳
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Hunan University
Original Assignee
Hunan University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Hunan University filed Critical Hunan University
Priority to CN202210882051.6A priority Critical patent/CN115439877A/en
Publication of CN115439877A publication Critical patent/CN115439877A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/103—Static body considered as a whole, e.g. static pedestrian or occupant recognition
    • 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
    • G06V10/443—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 matching or filtering
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
    • G06V10/75—Organisation of the matching processes, e.g. simultaneous or sequential comparisons of image or video features; Coarse-fine approaches, e.g. multi-scale approaches; using context analysis; Selection of dictionaries
    • G06V10/751—Comparing pixel values or logical combinations thereof, or feature values having positional relevance, e.g. template matching
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
    • G06V10/772—Determining representative reference patterns, e.g. averaging or distorting patterns; Generating dictionaries

Abstract

The invention discloses a human skeleton recognition method and system based on template matching, wherein a human skeleton model is constructed, and a scattered and sparse point set of coordinates to be recognized is obtained; based on length constraint of template edges in the human skeleton model and connection constraint of template points, screening out each template point, candidate matching real-time points and candidate matching real-time points corresponding to the template edges from the scattered sparse points of the coordinates to construct a matching dictionary; and traversing template edges and template points of the human skeleton model based on the matching dictionary, and identifying the human skeleton from the scattered and sparse points of the coordinates to be identified. The method determines the candidate matching real-time point and the candidate matching real-time edge by utilizing the length constraint of the template edge and the connection constraint of the template point, thereby constructing a matching dictionary, and then identifies the template edge and the template point edge by edge and point by point, thereby realizing the centralized identification of the three-dimensional human skeleton from the single-frame scattered sparse points.

Description

Human skeleton recognition method and system based on template matching
Technical Field
The invention relates to the field of computer vision, in particular to a human skeleton recognition method and system based on template matching.
Background
Pure visual three-dimensional human skeleton recognition has been an important research topic in the field of computer vision. The three-dimensional human body skeleton is an important analysis object for an intelligent system to understand images/videos with human as a center, intuitively represents a human body model in a form of a graph structure through parameterized joint point positions and limb parts, has the remarkable advantages of high robustness and high-density information expression capacity, and is a common human body modeling mode.
The human body sparse reconstruction method can obtain the three-dimensional coordinates of each mark point and store the three-dimensional coordinates in a digital form. However, these reconstruction markers are still disordered and only contain scattered sparse point sets of coordinate information, and a connection relationship between the sparse point sets and the human skeleton model needs to be constructed, so that the human body actions are also converted into digital representations, and this process is also called human skeleton recognition. The human skeleton is a sparse non-rigid body, and the internal graph structure has high uncertainty and variability, so that the identification of the human skeleton from scattered sparse point sets is a challenging task. The pure manual method for constructing the human body skeleton frame by frame obviously consumes quite time and is tedious, and errors are easy to occur, so that a method for automatically identifying the human body skeleton is very necessary to be researched, and key technologies needing key breakthroughs are as follows: the integral identification problem can be abstracted into a set of known template graphs and a scattered sparse point set, the problem that whether similar template graph subsets exist in a graph set formed by the scattered sparse point set or not is solved, and the integral identification can construct a matching relation between a single-frame scattered sparse point set and a human skeleton model;
aiming at the challenges of human skeleton recognition, commercial motion capture data analysis software manually constructs a human skeleton template, and then automatically recognizes the graphic structure of the human skeleton according to a predefined template during motion capture, so that the method effectively reduces manual labor. However, in the actual recognition process, the case of recognition failure still frequently occurs. Taking Cortex software of magic corporation of America as an example, the four human body skeleton recognition results of the same frame are different, the real-time tracking result of each human body skeleton node is also different, and the structure of the human body skeleton recognized by Cortex is not stable enough.
Disclosure of Invention
The invention provides a human body skeleton identification method and system based on template matching, which are used for solving the technical problem of poor robustness of the existing human body skeleton identification method.
In order to solve the technical problems, the technical scheme provided by the invention is as follows:
a human skeleton recognition method based on template matching comprises the following steps:
constructing a human skeleton model, and acquiring a coordinate scattered sparse point set to be identified;
based on length constraint of template edges and connection constraint of template points in the human skeleton model, screening out each template point, candidate matching real-time points and candidate matching real-time points corresponding to the template edges from the coordinate scattering sparse points in a centralized manner to construct a matching dictionary;
and traversing the template edges and template points of the human body skeleton model based on the matching dictionary, and identifying the human body skeleton from the scattered sparse points of the coordinates to be identified in a centralized manner.
Preferably, the length constraint of the template edge is: the length of the candidate matching edge of each template edge is within the length interval; the template point connection constraint is as follows: if the real time point
Figure BDA0003764550200000021
As template points
Figure BDA0003764550200000022
The candidate matching real-time point of (2), then the real-time point
Figure BDA0003764550200000023
Must be the template point
Figure BDA0003764550200000024
The candidates for all template edges connected match the common point of the real-time edge.
Preferably, the human skeleton model comprises a length interval of each template edge; based on the length constraint of each template edge and the connection constraint of each template point in the human skeleton model, screening out each template point, candidate matching real-time points and candidate matching real-time points corresponding to the template edges from the scattered sparse points of the coordinates to be identified, and the method comprises the following steps:
constructing a real-time edge set of the human skeleton to be identified by taking the sparse points in the coordinate scattering sparse point set as real-time points and taking a connecting line between any two real-time points as a real-time edge;
screening out the candidate matching real-time edge of each template edge from the real-time edge set based on the length of the candidate matching edge of each template edge in the length interval so as to construct the candidate matching real-time edge set of each template edge;
for each template point, constructing a candidate matching real-time point set of each template edge connected with the template point, wherein the candidate matching real-time point set of each template edge comprises real-time points connected with both ends of all candidate matching real-time edges of the template point; performing intersection operation on all candidate matching real-time point sets of the template edges connected with the template edges, and taking the real-time points obtained by the intersection operation as candidate matching real-time points of the template points;
and for each template edge, synthesizing the candidate matching real-time points of the common template points on the two sides of each template edge, and screening and updating the candidate matching real-time edges of each template edge.
Preferably, traversing the template edges and the template points of the human skeleton model based on the matching dictionary, and identifying the human skeleton from the scattered and sparse points of the coordinates to be identified, the method comprises the following steps:
step 1: when traversal starts, firstly, initialization is carried out, a template point Marker1 connected with one end of a template to be identified is selected as an identification starting point, a candidate matching real-time point is selected from a candidate matching real-time point set of the template point Marker1 as a matching point of the template point Marker1, and the other candidate matching real-time points and the current matching condition are pressed into a traversal stack;
step 2: judging whether any candidate matching real-time edge exists in the candidate matching real-time edge set of the template edge to be recognized from the candidate matching real-time edge set of the template point Marker2 at the other end connected with the template edge to be recognized, wherein the candidate matching real-time edge set of the template edge to be recognized meets the condition that one end is connected with the matching point of the template point Marker1, and the other end is connected with any candidate matching real-time point of the template point Marker 2; if the template point Marker1 does not exist, the identification of the template point Marker1 is failed, and the process of traversing the stack top of the stack is taken to continue the identification. If the candidate matching real-time edges meeting the conditions exist, the candidate matching real-time edges are used as the matching edges of the template edges to be identified, the next template edge is traversed until all the template edges are traversed, the connection relation of the real-time points can meet the connection relation determined by all the template edges, and then the whole template is identified.
Preferably, if more than one real-time edge meets the condition, one real-time edge is selected as a matching object of the current template edge to be continuously identified, and the rest real-time edges and the current identification result are stored in the traversal stack until the overall identification is successful.
Preferably, if the current matching precondition fails and the stored to-be-matched process in the stack is traversed, it indicates that the combination of all candidate matching real-time points and real-time edges cannot meet the target template structure, and at this time, it is determined that the overall recognition fails.
A computer system comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the steps of the method when executing the computer program.
The invention has the following beneficial effects:
1. according to the template matching-based human body skeleton identification method and system, the candidate matching real-time points and real-time edges are determined by utilizing the length constraint of the template edges and the connection constraint of the template points, so that a matching dictionary is constructed, the template edges and the template points are identified edge by edge and point by point, the three-dimensional human body skeleton is identified from a single-frame scattered sparse point set, and the method has good robustness on the number of marks and different action types.
In addition to the above-described objects, features and advantages, the present invention has other objects, features and advantages. The present invention will be described in further detail below with reference to the accompanying drawings.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this application, illustrate embodiments of the invention and, together with the description, serve to explain the invention and not to limit the invention. In the drawings:
FIG. 1 is a flow chart of a method of the present invention;
FIG. 2 is a candidate matching live-point plot of screened template points in the present invention.
Detailed Description
Embodiments of the invention will be described in detail below with reference to the drawings, but the invention can be implemented in many different ways as defined and covered by the claims.
The first embodiment is as follows:
the implementation discloses a human skeleton recognition method based on template matching, and the flow of the method comprises the following three stages.
Data preprocessing: the human skeleton template information comprises the total number n of template points, the names and the numbers of the template points, the total number m of template edges, the minimum length, the maximum length, the telescopic range (namely a length interval, wherein the minimum length is the lower limit of the length interval, and the maximum length is the upper limit of the length interval) of each template edge, and the numbers of two connected end points. In the following description, points in the template are referred to as template points, denoted as template points
Figure BDA0003764550200000031
It represents the ith template point. The edges in the template, referred to as template edges, are designated as template edges
Figure BDA0003764550200000032
It represents the u-th template edge. Each template edge defines a length expansion range. The change degree of the human skeleton structure is determined, and the larger the telescopic range is, the richer the change of the human skeleton structure is. During motion capture, the reconstructed sparse points are called real-time points and are written as
Figure BDA0003764550200000033
It represents the jth real-time point, and the connection between any two real-time points is called a real-time edge and is recorded as
Figure BDA0003764550200000034
It represents the v-th real-time edge, and the total number of the real-time edges is denoted as q. For the convenience of subsequent calculation, the known real-time point, real-time edge and human skeleton template information are preprocessed firstly. And sequencing all real-time edges in the order from small to large in length.
Constructing a matching dictionary: the matching dictionary records the candidate matching real-time point and the candidate matching real-time edge corresponding to each template point and template edge so as to identify the human skeleton template point by point edge by edge. The construction process of the matching dictionary comprises three steps of preliminarily screening the real-time edges of the candidate matching, screening the real-time points of the candidate matching and updating the real-time edges of the candidate matching.
Step 1: real-time edges having a length within the scalable length of the template edge are classified into a group and serve as candidate matching real-time edges for the template edge. The set of all candidate matching real-time edges is written as
Figure BDA0003764550200000041
Wherein
Figure BDA0003764550200000042
Representing edges of a template
Figure BDA0003764550200000043
The corresponding candidate matches the real-time edge. Firstly initializing a template edge serial number u and a real-time edge serial number v, namely, enabling u =1, v =1, then fixing the template edge serial number u, and sequentially searching for the edge positioned on the template
Figure BDA0003764550200000044
Real-time edges within a scalable range until the real-time edges exceed the template edges
Figure BDA0003764550200000045
A defined long edge value, at this time
Figure BDA0003764550200000046
The process of searching for the candidate match real-time edge is ended. Then continuously searching the candidate match of the next template edgeAnd repeating the previous operation until all candidate matching real-time edges of the template edges are aligned.
And 2, step: and selecting candidate matching real-time points corresponding to the template points. The real-time point of candidate matching corresponding to each template point is recorded
Figure BDA0003764550200000047
Wherein
Figure BDA0003764550200000048
Representing template points
Figure BDA0003764550200000049
The corresponding candidate match real time point. A real time point
Figure BDA00037645502000000410
If the matching can be successfully carried out to the corresponding template point
Figure BDA00037645502000000411
Then the template point
Figure BDA00037645502000000412
The candidate matching real-time edges of all connected template edges necessarily contain more than one common point, and the common points are the candidate matching real-time points of the template points.
And step 3: after the real-time points corresponding to the template points are basically determined, the corresponding relation between the points is utilized, and the matching relation between the template edges and the real-time edges is sequentially updated according to the arrangement sequence of the template edges.
Traversing and identifying the human skeleton according to the template: and sequentially identifying the matching relationship between the template point and the candidate matching real-time point according to the template edge. The method comprises the following two steps:
step 1: the traversal is initialized at first, and the matching condition of one end Marker1 of the template edges is determined. And taking an end point Marker1 with a smaller edge number of the current template as a first template point, namely a recognition starting point, if the candidate matching real-time point comprises a plurality of points, selecting one candidate matching real-time point as a current recognition result of the Marker1, and pressing the other candidate matching real-time points and the current matching condition into a traversal stack. And if the current recognition result of the Marker1 fails in subsequent recognition, continuously trying a new round of template recognition operation in the process of extracting the stack top of the traversal stack, and repeating the steps until the matching relation between all the template points and the real-time point is successfully established.
And 2, step: after the matching point of the end point Marker1 of the current template edge is determined, the matching point of another end point Marker2 connected with the current template edge is continuously calculated. If the Marker1 has not determined the matching point, the matching point is determined by an initialization method. And extracting a real-time edge containing a real-time point of a matching Marker1 from the candidate matching real-time edges of the template edge to be identified, wherein one end point of the real-time edge is a matching real-time point of the Marker1, and the other end point is a matching real-time point of the Marker 2. If the end point is not in the candidate matching real-time point of the Marker2, the identification of the Marker1 is failed, and the stack top process of the traversal stack is taken to continue the identification. If more than one real-time edge meets the condition, selecting one real-time edge as a matching object of the current template edge to continue to identify, and storing the other real-time edges and the current identification result into the traversal stack until the overall identification is successful. If the current matching precondition fails and the stored process to be matched is not in the traversal stack, it is indicated that all combinations of the candidate matching real-time points and the real-time edges cannot meet the target template structure, and at this time, the overall recognition is judged to fail. And if the connection relation of the real-time points can meet the connection relation determined by all the template edges after traversing all the template edges, the integral identification is considered to be successful.
In addition, in this embodiment, a computer system is also disclosed, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and when the processor executes the computer program, the steps of the method are implemented.
The human body skeleton recognition method based on template matching is a system, candidate matching real-time points and real-time edges of the template points are determined by utilizing length constraints of the template edges and connection constraints of the template points, so that a matching dictionary is constructed, the template edges and the template points are recognized edge by edge and point by point, the three-dimensional human body skeleton is recognized from a single-frame scattered sparse point in a centralized mode, and the method has good robustness on the number of marks and different action types.
Example two:
the second embodiment is the preferred embodiment of the first embodiment, and is different from the first embodiment in that a human skeleton recognition method based on template matching is introduced by combining specific data:
in this embodiment, a human skeleton recognition method based on template matching is disclosed, as shown in fig. 1.
S1, obtaining a scattered sparse point set;
a scattered sparse point set of a three-dimensional coordinate is obtained by a human body sparse reconstruction method and stored in a digital form. The number of scattered sparse points is 46, and the total number of real-time edges is
Figure BDA0003764550200000051
The information about the real-time point and the real-time side length is shown in table 1, and two end points of one side are represented by Marker1 and Marker 2.
Figure BDA0003764550200000052
Table 1 example real-time points, real-time edge lengths, and real-time edge numbers
S2, preprocessing data;
and storing template information according to the rule that the serial number of the edge endpoint Marker1 of the template is greater than that of the endpoint Marker2, and the serial numbers of the endpoint markers 1 are sorted from small to large. Record the template point set as
Figure BDA0003764550200000053
The set of all template edges is denoted as
Figure BDA0003764550200000061
All sets of real-time points are recorded as
Figure BDA0003764550200000062
The set of all real-time edges is denoted as
Figure BDA0003764550200000063
As shown in table 2, 5 template points were set.
Figure BDA0003764550200000064
TABLE 2 template information storage mode consisting of 5 template points
S3, constructing a matching dictionary;
as shown in fig. 1, the construction process of the matching dictionary comprises three steps of preliminarily screening the real-time edge of the candidate matching, screening the real-time point of the candidate matching, and updating the real-time edge of the candidate matching.
Step 1: template edge
Figure BDA0003764550200000065
Has a range of 140.69-143.92, and there are 5 real-time edges in the range, the detailed information of which is shown in table 3, and the three real-time edges are
Figure BDA0003764550200000066
Candidate matching real-time edge set of
Figure BDA0003764550200000067
Enter the subsequent screening process.
Figure BDA0003764550200000068
TABLE 3 template edge
Figure BDA0003764550200000069
Corresponding candidate matching real-time edge set
Figure BDA00037645502000000610
And 2, step: as shown in Table 2, the template points
Figure BDA00037645502000000611
The template edge comprises
Figure BDA00037645502000000612
The candidate matching real-time edges of the three edges are shown in table 4, and for simplification of representation, table 4 shows only the candidate matching real-time edges of part of the template edges, and the number of the candidate matching real-time edges in the actual situation is larger. Due to three template edges
Figure BDA00037645502000000613
There is a common endpoint
Figure BDA00037645502000000614
Therefore, if three template edges can be successfully identified, more than one common end points must exist in the three real-time edges matched with the template edges, and the common end points are template points
Figure BDA00037645502000000615
And adding these points to the real-time points of the candidate match
Figure BDA00037645502000000616
FIG. 2 shows screening of template spots
Figure BDA00037645502000000617
The set 1 contains real-time points representing template edges
Figure BDA0003764550200000071
All candidates of (2) match the end points of the real-time edge, set 2 represents
Figure BDA0003764550200000072
All candidates of (3) match the end points of the real-time edge, set 3 representing
Figure BDA0003764550200000073
All candidates of (2) match the end points of the real-time edge, see that only
Figure BDA0003764550200000074
Is the intersection of these three sets, i.e. exists and only
Figure BDA0003764550200000075
Matching real-time edges by candidates of three template edges
Figure BDA0003764550200000076
Shared (note that in practice it is the case that
Figure BDA0003764550200000077
The real time points of the candidate matching are shown in table 6, and only the results are shown here for convenience of explanation
Figure BDA0003764550200000078
The screening process) of (a), thus the real time point
Figure BDA0003764550200000079
Is confirmed as a template point
Figure BDA00037645502000000710
The candidate matching real time point of (2). Confirming the real time point
Figure BDA00037645502000000711
Is composed of
Figure BDA00037645502000000712
While the candidate matching real time point is being obtained, the updating can be continued
Figure BDA00037645502000000714
And the candidate matching real-time points of the other three template points on the three template sides.
Figure BDA00037645502000000715
TABLE 4 template edge
Figure BDA00037645502000000716
Corresponding candidate matching real-time edge setsClosing box
Figure BDA00037645502000000717
And step 3: template edge
Figure BDA00037645502000000718
The two end points are template points
Figure BDA00037645502000000719
And
Figure BDA00037645502000000720
after the classification and matching of the template points are completed, the corresponding relationship between the template points and the real-time points is shown in table 5, and the template points
Figure BDA00037645502000000721
Including
Figure BDA00037645502000000722
And
Figure BDA00037645502000000723
template point
Figure BDA00037645502000000724
Comprises
Figure BDA00037645502000000725
And
Figure BDA00037645502000000726
then to the template edge
Figure BDA00037645502000000727
One end point of the matched real-time edge should be a real-time point
Figure BDA00037645502000000728
Or
Figure BDA00037645502000000729
The other endpoint should be a real-time point
Figure BDA00037645502000000730
Or
Figure BDA00037645502000000731
And after screening, the edges of the template
Figure BDA00037645502000000732
The candidate matching real-time edges of (2) are shown in Table 4, which shows that only real-time edges exist
Figure BDA00037645502000000733
Meets the requirements, thereby the edges of the template
Figure BDA00037645502000000734
Can be further reduced to
Figure BDA00037645502000000735
Similar operations are performed on other template edges, and the updated candidate matching real-time edges of the 8 template edges are shown in table 6, which form a matching dictionary in the subsequent matching process
Figure BDA00037645502000000736
TABLE 5 set of candidate matching real-time points G corresponding to template points p
S4, traversing and identifying according to template edges;
step 1: the real-time point of candidate matching for each template point is shown in Table 5, where the template points
Figure BDA0003764550200000081
The candidate matching real-time points are respectively
Figure BDA0003764550200000082
Selecting real time points
Figure BDA0003764550200000083
As template points
Figure BDA0003764550200000084
The matching points are subjected to subsequent identification, other candidate matching real-time points are stored in a stack for temporary reservation, if the identification failure occurs in the subsequent template points, the template points return to the position, and the real-time points are continuously taken
Figure BDA0003764550200000085
As template points
Figure BDA0003764550200000086
And carrying out subsequent identification on the matching points until the overall identification is successful.
Step 2: the candidate matching real-time edges for each template edge are shown in Table 6 at the template point
Figure BDA0003764550200000087
The matching point of (2) is a real-time point
Figure BDA0003764550200000088
Under the premise of (2), calculating the current endpoint Marker2 (namely the template point)
Figure BDA0003764550200000089
) Matching the real-time points. Template edge
Figure BDA00037645502000000822
Has only one candidate matching edge
Figure BDA00037645502000000810
With an endpoint of
Figure BDA00037645502000000811
And
Figure BDA00037645502000000812
apparently not including the real-time point
Figure BDA00037645502000000813
To illustrate the point of real time
Figure BDA00037645502000000814
And template point
Figure BDA00037645502000000815
Corresponding to a precondition error. Continuing the process of extracting the top of the stack until the top of the stack is successfully extracted
Figure BDA00037645502000000816
As template points
Figure BDA00037645502000000817
Matching points of, at the time of, real time
Figure BDA00037645502000000818
Will be used as template points
Figure BDA00037645502000000819
The matching points are identified subsequently, and the template edges are obtained
Figure BDA00037645502000000820
I.e. the identification is successful.
Figure BDA00037645502000000821
Figure BDA0003764550200000091
TABLE 6 optimized candidate matching real-time edge set G corresponding to template edges e
The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention, and various modifications and changes may be made by those skilled in the art. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims (7)

1. A human skeleton recognition method based on template matching is characterized by comprising the following steps:
constructing a human skeleton model, and acquiring a coordinate scattered sparse point set to be identified;
based on length constraint of template edges in the human skeleton model and connection constraint of template points, screening out each template point, candidate matching real-time points and candidate matching real-time points corresponding to the template edges from the coordinate scattering sparse points in a centralized manner to construct a matching dictionary;
and traversing the template edges and template points of the human body skeleton model based on the matching dictionary, and identifying the human body skeleton from the scattered and sparse points of the coordinates to be identified.
2. The template matching-based human skeleton recognition method according to claim 1, wherein the length constraints of the template edges are: the length of the candidate matching edge of each template edge is within the length interval; the template point connection constraint is as follows: if the real time point
Figure FDA0003764550190000011
As template points
Figure FDA0003764550190000012
The candidate matching real-time point of (2), then the real-time point
Figure FDA0003764550190000013
Must be the template point
Figure FDA0003764550190000014
The candidates for all template edges connected match the common point of the real-time edge.
3. The template matching-based human body skeleton recognition method according to claim 1 or 2, wherein the human body skeleton model comprises a length interval of each template edge; based on the length constraint of each template edge and the connection constraint of each template point in the human skeleton model, screening out each template point, candidate matching real-time points and candidate matching real-time points corresponding to the template edges from the scattered sparse points of the coordinates to be identified, and the method comprises the following steps:
constructing a real-time edge set of the human skeleton to be identified by taking sparse points in the scattered coordinate sparse point set as real-time points and taking a connecting line between any two real-time points as a real-time edge;
screening out the candidate matching real-time edges of each template edge from the real-time edge set based on the length of the candidate matching edges of each template edge within the length interval of the candidate matching edges so as to construct a candidate matching real-time edge set of each template edge;
for each template point, constructing a candidate matching real-time point set of each template edge connected with the template point, wherein the candidate matching real-time point set of each template edge comprises real-time points connected with both ends of all candidate matching real-time edges of the template point; performing intersection operation on all candidate matching real-time point sets of the template sides connected with the template sides, and taking the real-time points obtained by the intersection operation as the candidate matching real-time points of the template points;
and for each template edge, synthesizing the candidate matching real-time points of the common template points on the two sides of each template edge, and screening and updating the candidate matching real-time edges of each template edge.
4. The template matching-based human body skeleton recognition method according to claim 3, wherein template edges and template points of the human body skeleton model are traversed based on the matching dictionary, and a human body skeleton is recognized from a scattered sparse point set of coordinates to be recognized, comprising the following steps:
step 1: when traversal starts, firstly, initializing, selecting a template point Marker1 connected with one end of a template to be identified as an identification starting point, selecting a candidate matching real-time point from a candidate matching real-time point set of the template point Marker1 as a matching point of the template point Marker1, and pressing the other candidate matching real-time points and the current matching condition into a traversal stack;
step 2: judging whether any candidate matching real-time edge exists in the candidate matching real-time edge set of the template edge to be recognized from the candidate matching real-time edge set of the template point Marker2 at the other end connected with the template edge to be recognized, wherein the candidate matching real-time edge set of the template edge to be recognized meets the condition that one end is connected with the matching point of the template point Marker1, and the other end is connected with any candidate matching real-time point of the template point Marker 2; if the template point Marker1 does not exist, the identification of the template point Marker1 is failed, and the process of traversing the stack top of the stack is taken to continue the identification. If the candidate matching real-time edges meeting the conditions exist, the candidate matching real-time edges are used as the matching edges of the template edges to be identified, the next template edge is traversed until all the template edges are traversed, the connection relation of the real-time points can meet the connection relation determined by all the template edges, and then the whole template is identified.
5. The template matching-based human body skeleton recognition method according to claim 4, wherein if more than one real-time edge meets the condition, one real-time edge is selected as a matching object of the current template edge to continue recognition, and the rest real-time edges and the current recognition result are stored in a traversal stack until the overall recognition is successful.
6. The template matching-based human skeleton recognition method according to claim 5, wherein if the current matching precondition fails and the to-be-matched process which is not stored in the stack is traversed, it is indicated that the combination of all candidate matching real-time points and real-time edges cannot meet the target template structure, and at this time, it is determined that the overall recognition fails.
7. A computer system comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the steps of the method of any one of claims 1 to 6 are performed when the computer program is executed by the processor.
CN202210882051.6A 2022-07-26 2022-07-26 Human skeleton recognition method and system based on template matching Pending CN115439877A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210882051.6A CN115439877A (en) 2022-07-26 2022-07-26 Human skeleton recognition method and system based on template matching

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210882051.6A CN115439877A (en) 2022-07-26 2022-07-26 Human skeleton recognition method and system based on template matching

Publications (1)

Publication Number Publication Date
CN115439877A true CN115439877A (en) 2022-12-06

Family

ID=84240859

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210882051.6A Pending CN115439877A (en) 2022-07-26 2022-07-26 Human skeleton recognition method and system based on template matching

Country Status (1)

Country Link
CN (1) CN115439877A (en)

Similar Documents

Publication Publication Date Title
EP0534446B1 (en) System with approximation means for recognizing graphical elements in a drawing
US6118897A (en) Interactive drawing recognition processing method and apparatus thereof
US9348877B2 (en) Methods, systems, and data structures for performing searches on three dimensional objects
US20040153824A1 (en) System and method for creating and updating a three-dimensional model and creating a related neutral file format
CN1136848A (en) Fingerprint matching system
Tanase et al. Polygon decomposition based on the straight line skeleton
CN116630990B (en) RPA flow element path intelligent restoration method and system
CN115439877A (en) Human skeleton recognition method and system based on template matching
CN112948911A (en) System and method for converting S3D bridge model into MapGIS model
CN110334237B (en) Multi-mode data-based three-dimensional object retrieval method and system
CN110659348A (en) Group enterprise universe risk fusion analysis method and system based on knowledge reasoning
JP5061147B2 (en) Image search device
CN110110024B (en) Method for importing high-capacity VCT file into spatial database
CN110019830B (en) Corpus processing method, corpus processing device, word vector obtaining method, word vector obtaining device, storage medium and equipment
CN101807244A (en) Machine recognition and reconstruction method
CN116258842B (en) Fingerprint template dynamic splicing optimization system and method
CN117370592B (en) Part similarity recognition method based on machine learning
CN116306574B (en) Big data mining method and server applied to intelligent wind control task analysis
CN117194410B (en) Method and system for generating business report by artificial intelligence language model
CN112560273B (en) Method and device for determining execution sequence of model components facing data flow model
JPH0614360B2 (en) Object recognition device
Angeles-Yreta et al. Computing similarity among 3D objects using dynamic time warping
US6625330B1 (en) Method for controlled and meaningful shape modifications
JPH01130278A (en) Image information retrieving system
CN112115141A (en) Data lineage analysis method based on graph database

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