Human body posture database image retrieval method and system
Through key point representation and two-way clamping search strategies, the retrieval efficiency and accuracy of human posture recognition in complex backgrounds are solved, the efficiency and accuracy of posture retrieval is improved, the manual intervention is reduced, and the degree of automation is enhanced.
Patent Information
- Application Number
- CN202411093312.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-09
- Publication Date
- 2024-11-22
- Estimated Expiration
- 2044-08-09
AI Technical Summary
In the complex image background, the search efficiency and accuracy of human posture recognition are affected, especially in the field of safety monitoring. Factors such as human occlusion increase the background complexity, resulting in a decrease in the success rate of effective retrieval, and the training complexity of artificial intelligence models Exponential growth.
The key point representation method is constructed through the approximate location distribution of key points, and a two-way clamping search strategy is adopted, combining the confidence and spatial relationship of key points, and conducting accurate and loose-fitting searches to reduce interference from complex image backgrounds and improve posture retrieval efficiency.
Improves the efficiency, accuracy and robustness of pose retrieval, reduces manual intervention, enhances the degree of automation, and can accurately match pose samples in complex environments.
Smart Images

Figure CN119003808A_ABST
Abstract
Description
[Technical field]
[0001] The present invention belongs to the technical field of posture retrieval based on artificial intelligence, and in particular relates to a method and system for image retrieval of a human body posture database. [Background technology]
[0002] Posture recognition technology is a human posture analysis method based on computer vision. It extracts the key points and skeleton information of the human body by analyzing the human body's posture, and models and recognizes the human body's posture. With the development of deep learning technology, posture recognition technology has been widely used and studied. Compared with other biometric recognition technologies, the advantages of posture recognition are non-contact, non-invasive, easy to perceive, and difficult to hide and disguise the target. In addition, posture analysis can easily distinguish different behavior patterns of individual people, such as walking, running, or carrying heavy objects. Posture recognition is an important research direction in the field of computer vision, which aims to detect and identify human postures in images or videos. Human posture information can be used to further perform tasks such as action recognition, human-computer information interaction, and abnormal behavior detection. In practical applications, posture recognition technology is widely used in intelligent monitoring, human-computer interaction, virtual reality, motion analysis and other fields. In the field of motion analysis, posture recognition can be used to monitor and evaluate athletes' movements in real time to improve the effect of sports training. In the field of virtual reality, posture recognition technology allows users to perform natural body movements in a virtual environment, improving the immersion and authenticity of virtual reality. Gesture recognition is particularly suitable for access control systems, security monitoring, human-machine exchange, medical diagnosis and other fields. In the field of security, gesture recognition technology has a wide range of applications and economic value, and can effectively monitor and prevent criminal behavior.
[0003] However, on the one hand, with the diversification of application scenarios, the complexity of the environment where the human body is located will greatly increase the complexity of the image background, especially in the field of security monitoring, factors such as human body occlusion further increase the complexity of the background in the image, greatly reducing the success rate of effective retrieval. On the other hand, there is a close relationship between artificial intelligence models and gesture recognition. Artificial intelligence models, especially deep neural network models, are widely used in various steps of gesture recognition. Artificial intelligence models for specific scenarios need to be trained specifically for specific vertical applications, and special artificial intelligence models are set to integrate the analysis results of artificial intelligence models used in various processing steps such as image analysis and image segmentation. The superposition of the use of artificial intelligence models causes an exponential increase in complexity; and the accuracy of training is affected by various environmental factors, which will greatly increase the complexity of calculation; Based on the above problems, the present invention constructs a key point representation method through the approximate position distribution of key points, performs two-way clamping retrieval with different optimization indicators, and reduces the retrieval interference caused by complex image backgrounds, and improves the efficiency of gesture retrieval. [Summary of the invention]
[0004] In order to solve the above problems in the prior art, the present invention proposes a human body posture database image retrieval method and system, the method comprising:
[0005] Step S1: Segment the image into three types: human body area, non-human body area and fuzzy area; wherein the fuzzy area is the area between the human body area and the non-human body area;
[0006] Step S2: extracting human features and locating key points;
[0007] Step S3: Classify the key points into multiple levels based on the key point confidence;
[0008] Step S4: searching a human body posture database based on key points to determine a posture sample pool; determining a target posture sample from the posture sample pool by adopting a two-way clamping method based on the positional relationship between the key points;
[0009] The step S4 specifically includes the following steps:
[0010] Step S41: Taking the specific position as the center point, the segmentation result including the human body area and the fuzzy area is divided into N sub-areas; the key points falling into each sub-area are determined, and the current N-ary identifier (cls n ), cls n =(p nm ), where: p nm The key point type or the position of the key point in the image / segmentation result; the element cls n It is M nThe set of elements contains the M in the nth subregion n key points; m is cls n The element number in; m = 1 ~ M n ;M n It is cls n The number of elements in ;
[0011] Step S42: based on the current N-gram identifier, use the key point type and its distributed sub-area to retrieve the human posture database to obtain a posture sample pool; specifically: put posture samples whose number of elements of the N-gram identifier in the human posture database is equal to the number of elements in the current N-gram identifier and whose key point types are the same in each element in the identifier into the posture sample pool;
[0012] Step S43: determining the target posture sample from the posture sample pool; specifically: selecting the key point with the highest level to form the initial value of the left key point set; setting the initial value of the right key point set to be equal to the key point set composed of all key points; retrieving the posture sample pool from the left key point set and the right key point set respectively, performing two-way clamping until the retrieval result meets the convergence condition, and obtaining the target posture sample; taking the posture of the target posture sample as the retrieval result;
[0013] The step S43 specifically includes the following steps:
[0014] Step S431: Determine the sub-region where each key point in the left key point set falls, and the key points falling in the sub-region n constitute the nlth left key point set cls nl ; Build its identity (cls nl );
[0015] Step S432: Based on each nlth left key point set cls nl The positional relationship between each element E1 and any two other elements E2 and E3 in the left key point set is used to select the posture samples that meet the conditions from the posture sample pool and put them into the left posture sample set to be selected; the condition is that there are elements in the posture sample with the same positional relationship as E1, E2, E3;
[0016] Step S433: searching the pose sample pool based on all key point types and their positions in the right key point set in the image, and if there is a pose sample that matches all key point types and their positions, putting it into the right pose sample set to be selected;
[0017] Step S434: if the posture samples in the left posture sample set to be selected and the right posture sample set to be selected have an intersection, the posture samples in the intersection are used as target posture samples;
[0018] Step S435: Update the left key point set and the right key point set; if the left key point set and the right key point set cannot be updated or the left key point set and the right key point set are the same, proceed to the next step; otherwise, return to step S431;
[0019] Step S44: taking the posture of the target posture sample as the retrieval result.
[0020] Furthermore, an artificial intelligence model is used to segment the image based on three categories.
[0021] Furthermore, a deep neural network based on semantic segmentation is used to recover the category of each pixel from the abstract features of the convolutional layer.
[0022] Furthermore, the human body image comes from video surveillance, professional motion capture equipment or static pictures.
[0023] Furthermore, before executing step S1, the human body image is preprocessed.
[0024] Furthermore, morphological operations are performed on the segmentation results.
[0025] Furthermore, the morphological operation includes removing small objects and / or filling holes.
[0026] A human body posture database image retrieval server comprises a processor, the processor is coupled to a memory, the memory stores program instructions, and when the program instructions stored in the memory are executed by the processor, the human body posture database image retrieval method is implemented.
[0027] A human body posture database image retrieval system, the system is used to implement the human body posture database image retrieval method.
[0028] A computer-readable storage medium includes a program, which, when running on a computer, enables the computer to execute the human body posture database image retrieval method.
[0029] The beneficial effects of the present invention include:
[0030] (1) The key point representation method is constructed through the approximate position distribution of the key points, and different optimization indicators are used to perform precise and loose matching bidirectional clamp retrieval, which reduces the retrieval interference caused by complex image background and improves the efficiency, accuracy and robustness of posture retrieval;
[0031] (2) The influence of complex environmental factors is used in the process of determining the confidence of key points, which provides a quantitative reliability index for the positioning of key points. Based on the classification of key point confidence and the spatial relationship between key points, a two-way clamping search strategy is used to start from a set of key points with high confidence and gradually approach a set of key points with low confidence until the target pose sample is found. This can improve the search efficiency and accurately retrieve the sample that best matches the query pose from the pose sample pool.
[0032] Furthermore, by considering the inner angles between key points, this method can perform robust matching on pose samples and obtain accurate retrieval results even when there are certain errors in the key point positions.
[0033] (3) The construction of N-ary identifiers provides a flexible way to organize and retrieve key points, making it easier to build a sample pool containing multiple postures, providing rich sample resources for subsequent precise retrieval. The value of N can also be adjusted according to different application requirements to obtain better retrieval performance, thereby reducing manual intervention and improving the degree of automation of posture retrieval.
Brief Description of the Drawings
[0034] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of the present application, but do not constitute an improper limitation of the present invention. In the drawings:
[0035] Figure 1 This is a schematic diagram of the human body posture database image retrieval method provided by the present invention. [Specific implementation method]
[0036] The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments, wherein the exemplary embodiments and descriptions are only used to explain the present invention, but are not intended to limit the present invention.
[0037] The present invention proposes a human body posture database image retrieval method and system, as shown in the attached Figure 1 As shown, the method comprises the following steps:
[0038] Step S1: Segment the image into three types: human body area, non-human body area and fuzzy area; wherein the fuzzy area is the human body image area between the human body area and the non-human body area;
[0039] Preferably: an artificial intelligence model is used to segment the image based on three categories, the output of the artificial intelligence model is the probability that each pixel belongs to a human body, pixels with a probability greater than a high confidence threshold are human body pixels, pixels with a probability less than a low confidence threshold are non-human body pixels, and pixels within the confidence threshold are human body segmentation results;
[0040] Preferred: Use a deep neural network based on semantic segmentation to recover the category of each pixel from the abstract features of the convolutional layer;
[0041] Preferred: Post-process the segmentation results, such as morphological operations, removing small objects, filling holes, etc., to improve the segmentation results;
[0042] Preferably: the human body image is from video surveillance, professional motion capture equipment or static pictures; alternatively: the image is accurately annotated to distinguish between human body areas, non-human body areas and fuzzy areas. This usually requires manual operation or the use of semi-automatic tools to improve efficiency;
[0043] Preferred: Preprocess the human body image, such as resizing, normalizing brightness and contrast, to reduce the influence of factors such as lighting and scale;
[0044] Step S2: extracting human features and locating key points;
[0045] The human body feature extraction is specifically: extracting human body features from the image; the human body features include edge features and texture features;
[0046] Preferred: Use the network structure commonly used in deep learning to extract human features, such as convolutional neural network and recurrent neural network;
[0047] The key point positioning is specifically performed by: using a key point determination model to locate key points based on human body features; for example: based on the extracted human body features, using a key point determination model based on linear regression, support vector machine algorithm to predict the positions of various key points of the human body; or using a key point determination model based on a traditional image analysis method to locate key points;
[0048] Preferably: the key points include multiple types located at various parts of the human body; when there are multiple key points of the same type, the key point with the highest confidence is selected, and the key point is subsequently graded;
[0049] Alternative: When there are multiple key points of the same type, select the key points with the highest confidence and perform subsequent level classification on them;
[0050] Preferred: For example: the center of the head, the nose or eyes, the back of the neck where the head and body connect, the tops of both shoulders, the elbows of both arms, the locations of both wrists, the hands, the sides of the hips, the locations of both knees, the locations of the ankles, the bottoms of the feet or the locations of the heels, the locations of both eyes, the top or center of the ears, the rotation points of the shoulders, the rotation points of the hips, the rotation points of the elbows, the rotation points of the wrists, the rotation points of the knees, and / or the rotation points of the ankles;
[0051] Step S3: determining the confidence of the key points, and dividing the key points into multiple levels based on the confidence of the key points; specifically: dividing the key points into multiple levels according to the confidence interval into which the confidence of each key point output by the key point determination model falls; the greater the confidence, the higher the level;
[0052] Preferably: the complete confidence interval is divided so that the confidence of the key point falls evenly into the confidence interval obtained after the division;
[0053] Alternatively: the confidence interval is divided into three levels: high, medium, and low. According to the confidence score of each key point, it is assigned to the corresponding level: key points above the high confidence threshold are classified as high level; key points below the medium confidence threshold and above the low confidence threshold are classified as medium level; key points below the low confidence threshold are classified as low level;
[0054] Alternatively: the confidence of the key point is determined based on the area type and area involved in the human features used in the key point positioning process, and the key point is divided into multiple levels according to the confidence interval into which the confidence falls; specifically: the confidence fa of the key point is determined using the following formula (1); wherein: α is the confidence coefficient, cb, ncb, uncb are the human area, fuzzy area and non-human area areas involved in the key point positioning process respectively;
[0055] fa=(∑cb×1+ncb×α+uncb×1) / (cb+ncb+uncb) (1)
[0056] Preferred: α=0.98; by adjusting the confidence coefficient α, the model can adapt to different application scenarios and the accuracy of human body segmentation, thereby improving the adaptability of the model;
[0057] Alternatively: set α to the accuracy of human segmentation;
[0058] Preferably: when the key point is a three-dimensional coordinate, projection is performed to convert the three-dimensional key point into a two-dimensional key point;
[0059] Step S4: searching a human body posture database based on key points to determine a posture sample pool; determining a target posture sample from the posture sample pool by adopting a two-way clamping method based on the positional relationship between the key points;
[0060] The step S4 specifically includes the following steps:
[0061] Step S41: Taking the specific position as the center point, the segmentation result including the human body area and the fuzzy area is divided into N sub-areas; the key points falling into each sub-area are determined, and the current N-ary identifier (cls n ), clsn =(p nm ), where: p nm The key point type or the position of the key point in the human image; the element cls n It is M n The set of elements contains the M in the nth subregion n key points; m is cls n The element number in; m = 1 ~ M n ;M n It is cls n The number of elements; that is, the number of key points NM in each sub-region is the same or different; for example: 4-element identifier ((p1, p5)(p3)(p4)
[0062] (p2)), the left eye p1 and the right eye center point p5 are located in the first sub-area; ((p1, p5)(p3)(p4)(p2));
[0063] Preferably: the specific position is a central position;
[0064] Preferably: N is equal to 4 to 16; rays are uniformly constructed from the central position to divide the human body area into N sub-areas;
[0065] Preferably, the human body region division method is a preset division method, for example, when N=4, a two-dimensional coordinate axis is constructed based on the center position to divide the human body region into four sub-regions;
[0066] Step S42: Based on the current N-gram identifier, the human body posture database is retrieved using the key point type to obtain a posture sample pool; specifically, the posture samples whose number of elements of the N-gram identifier in the human body posture database is equal to the number of elements in the current N-gram identifier and whose key point types are the same in each element in the identifier are put into the posture sample pool;
[0067] Preferably: the N-ary identifier is set as an index;
[0068] Alternative: Use the entire human posture database as a posture sample pool;
[0069] Step S43: determining the target posture sample from the posture sample pool; specifically: selecting the key point with the highest level to form the initial value of the left key point set; setting the initial value of the right key point set to be equal to the key point set composed of all key points; retrieving the posture sample pool from the left key point set and the right key point set respectively, performing two-way clamping until the retrieval result meets the convergence condition, and obtaining the target posture sample; taking the posture of the target posture sample as the retrieval result;
[0070] The step S43 specifically includes the following steps:
[0071] Step S431: Determine the sub-region where each key point in the left key point set falls, and the key points falling in the sub-region n constitute the nlth left key point set cls nl ; Build its identity (cls nl );
[0072] Step S432: Based on each nlth left key point set cls nl The positional relationship between each element in and any two other elements in the left key point set is used to select the posture samples that meet the conditions from the posture sample pool and put them into the left posture sample set to be selected;
[0073] Preferably: the condition is that there are elements in the posture sample with the same position relationship as E1, E2, and E3;
[0074] Preferably: the condition is that there are elements in the posture sample that are of the same type as the key point of E1 and have the same positional relationship as E1 E2, E3;
[0075] Step S4321: Select an unprocessed posture sample from the posture sample pool as the current posture sample; and set the left key point set cls nl The initial value of number nl is 1;
[0076] Preferably: if there is no unprocessed gesture sample, then end step S432;
[0077] Step S4322: From the nlth left key point set cls nl Select an unprocessed key point p from nlc , if a key point p can be selected from any two other left key point sets nl1 and nl2 nl1 , p nl2 ; Make sure that there is a posture sample (cls) in the posture sample pool that satisfies condition (2) ns ), so that condition (3) is satisfied, then the nlth left key point set cls is added and updated nl The proportion or number of times that the key points meet the conditions; where: θ(a, b, c) is the inner angle between the straight line formed by the key points a and b and the straight line formed by the key points b and c;
[0078]
[0079] Step S4323: determine whether there are still unprocessed key points. If yes, return to step S4322; otherwise, if the proportion or number of times that meet the conditions is greater than a preset proportion or preset number of times, proceed to the next step; otherwise, continue to re-execute step S4321 to determine other posture samples in the posture sample pool;
[0080] Step S4324: If nl is greater than N, proceed to the next step; otherwise, set nl=nl+1 and return to step S4322;
[0081] Step S4325: putting the current posture sample into the left set of posture samples to be selected;
[0082] Alternatively, the step S432 is specifically as follows: if there is a posture sample (cls) in the posture sample pool that satisfies conditions (4) and (5) nb ), so that condition (6) is satisfied; then the posture sample is placed in the left set of posture samples to be selected; where: θ(a, b, c) is the inner angle between the straight line formed by key points a and b and the straight line formed by key points b and c;
[0083] in:
[0084] in:
[0085]
[0086] Alternatively, the step S432 is specifically as follows: if there is a posture sample and its identifier (cls nb ), so that condition (7) is satisfied; then the posture sample is placed in the left set of posture samples to be selected;
[0087] Where: θl1, θl2∈θ(pnb1, pnb2, pnb3)orθ(pnb1, pnb3, pnb2)orθ(pnb2, pnb3, pnb1); θs1, θs2∈(pnl1, pnl2, pnl3)orθ(pnl1, pnl3, pnl2)orθ(pnl2, pnl3, pnl1) (7);
[0088] Preferably: ≈ in formula (3) (7) means that the angles are roughly equal, and a certain deviation threshold can be used to define the agreed stem;
[0089] Step S433: searching the pose sample pool based on all key point types and their positions in the right key point set, and if there are pose samples that match all key point types and their positions, putting them into the right pose sample set to be selected; that is, finding pose samples with the same key point types and their position distributions;
[0090] Preferably: obtaining a posture sample in the posture sample pool, fixing the positions of a key point in the right key point set and the corresponding key points in the posture samples of the same type in the image; performing overall scaling and rotation on the other key points to determine whether the positions of the other key points in the image overlap or approximately overlap, and if so, determining that the one posture sample is matched;
[0091] Step S434: if the posture samples in the left posture sample set to be selected and the right posture sample set to be selected have an intersection, the posture samples in the intersection are used as target posture samples;
[0092] Step S435: Update the left key point set and the right key point set; if the left key point set and the right key point set cannot be updated or the left key point set and the right key point set are the same, proceed to the next step; otherwise, return to step S431;
[0093] The updating of the left key point set specifically comprises: selecting a key point with the highest level from the key points in the key point set that do not belong to the current left key point set and adding it to the left key point set;
[0094] The updating of the right key point set is specifically: deleting the key point with the lowest level in the right key point set; then the search condition based on the right key point set is always being relaxed; and then the search condition based on the left key point set is always being tightened; and both are search optimizations performed from different dimensions;
[0095] Alternatively: delete a key point with the lowest level in the right key point set;
[0096] Step S44: taking the posture of the target posture sample as the retrieval result; if there is no target posture sample, the retrieval fails;
[0097] Based on the same inventive concept, the present invention also provides a human body posture database image retrieval system, the system is used to implement the above-mentioned human body posture database image retrieval method;
[0098] Based on the same inventive concept, the present invention also provides a human body posture database image retrieval server, the server is used to implement the above-mentioned human body posture database image retrieval method;
[0099] Based on the same inventive concept, the present invention also provides a human body posture database image retrieval device,
[0100] The device is used to implement the above-mentioned human body posture database image retrieval method;
[0101] Based on the same inventive concept, the present invention also provides a human posture database image retrieval platform.
[0102] The platform is used to implement the above-mentioned human posture database image retrieval method;
[0103] A computer program (also referred to as a program, software, software application, script, or code) can be written in any form of a programming language, including an assembled or interpreted language, a declarative or procedural language, and it can be deployed in any form, including as a stand-alone program or as a module, component, subroutine, object, or other unit suitable for use in a computing environment. A computer program may, but need not, correspond to a file in a file system. A program can be stored in a portion of a file that holds other programs or data (e.g., one or more scripts stored in a markup language document), in a single file dedicated to the program, or in multiple collaborative files (e.g., files storing one or more modules, subroutines, or code portions). A computer program can be deployed to execute on one computer or on multiple computers located at one site or distributed across multiple sites and interconnected by a communication network.
[0104] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0105] The present invention is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of the processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0106] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1A function specified in one or more boxes.
[0107] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.
[0108] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.
Claims
1. A human body posture database image retrieval method, characterized in that: The method comprises: Step S1: Segment the image into three types: human body area, non-human body area and fuzzy area; wherein the fuzzy area is the area between the human body area and the non-human body area; Step S2: extracting human features and locating key points; Step S3: Classify the key points into multiple levels based on the key point confidence; Step S4: searching a human body posture database based on key points to determine a posture sample pool; determining a target posture sample from the posture sample pool by adopting a two-way clamping method based on the positional relationship between the key points; The step S4 specifically includes the following steps: Step S41: Taking the specific position as the center point, the segmentation result including the human body area and the fuzzy area is divided into N sub-areas; the key points falling into each sub-area are determined, and the current N-ary identifier (cls n ), cls n =(p nm ), where: p nm The key point type or the position of the key point in the image / segmentation result; the element cls n It is M n The set of elements contains the M in the nth subregion n key points; m is cls n The element number in; m = 1 ~ M n ;M n It is cls n The number of elements in ; Step S42: based on the current N-gram identifier, use the key point type and its distributed sub-area to retrieve the human posture database to obtain a posture sample pool; specifically: put posture samples whose number of elements of the N-gram identifier in the human posture database is equal to the number of elements in the current N-gram identifier and whose key point types are the same in each element in the identifier into the posture sample pool; Step S43: determining the target posture sample from the posture sample pool; specifically: selecting the key point with the highest level to form the initial value of the left key point set; setting the initial value of the right key point set to be equal to the key point set composed of all key points; retrieving the posture sample pool from the left key point set and the right key point set respectively, performing two-way clamping until the retrieval result meets the convergence condition, and obtaining the target posture sample; taking the posture of the target posture sample as the retrieval result; The step S43 specifically includes the following steps: Step S431: Determine the sub-region where each key point in the left key point set falls, and the key points falling in the sub-region n constitute the nlth left key point set cls nl ; Build its identity (cls nl ); Step S432: Based on each nlth left key point set cls nl The positional relationship between each element E1 and any two other elements E2 and E3 in the left key point set is used to select the posture samples that meet the conditions from the posture sample pool and put them into the left posture sample set to be selected; the condition is that there are elements in the posture sample with the same positional relationship as E1, E2, E3; Step S433: searching the pose sample pool based on all key point types and their positions in the right key point set in the image, and if there is a pose sample that matches all key point types and their positions, putting it into the right pose sample set to be selected; Step S434: if the posture samples in the left posture sample set to be selected and the right posture sample set to be selected have an intersection, the posture samples in the intersection are used as target posture samples; Step S435: Update the left key point set and the right key point set; if the left key point set and the right key point set cannot be updated or the left key point set and the right key point set are the same, proceed to the next step; otherwise, return to step S431; Step S44: taking the posture of the target posture sample as the retrieval result.
2. The human body posture database image retrieval method according to claim 1, characterized in that: Use artificial intelligence models to segment images based on three categories.
3. The human body posture database image retrieval method according to claim 2, characterized in that: A deep neural network based on semantic segmentation is used to recover the category of each pixel from the abstract features of the convolutional layer.
4. The human body posture database image retrieval method according to claim 3, characterized in that: The human body image comes from video surveillance, professional motion capture equipment or static pictures.
5. The human body posture database image retrieval method according to claim 4, characterized in that: Before executing step S1, the human body image is preprocessed.
6. The human body posture database image retrieval method according to claim 5, characterized in that: Perform morphological operations on the segmentation results.
7. The human body posture database image retrieval method according to claim 6, characterized in that: The morphological operations include removing small objects and / or filling holes.
8. A human body posture database image retrieval server, characterized in that: The method comprises a processor, the processor is coupled to a memory, the memory stores program instructions, and when the program instructions stored in the memory are executed by the processor, the human body posture database image retrieval method described in any one of claims 1 to 5 is implemented.
9. A human body posture database image retrieval system, characterized in that: The system is used to implement the human body posture database image retrieval method described in any one of claims 1-5.
10. A computer-readable storage medium, characterized in that: The invention comprises a program which, when running on a computer, enables the computer to execute the human body posture database image retrieval method described in any one of claims 1 to 5.
Citation Information
Patent Citations
Method for segmenting and indexing scenes by combining captions and video image information
CN101719144A
Image retrieval method and related equipment
CN114168768A
Face picture storage and retrieval method and device for cloud federation
CN114764447A
Image retrieval server and system, related retrieval and troubleshooting method
WO2017107866A1
