CN111639533A

CN111639533A - Gait feature-based posture detection method, device, equipment and storage medium

Info

Publication number: CN111639533A
Application number: CN202010350100.2A
Authority: CN
Inventors: 田金戈; 徐国强
Original assignee: OneConnect Financial Technology Co Ltd Shanghai
Current assignee: OneConnect Financial Technology Co Ltd Shanghai
Priority date: 2020-04-28
Filing date: 2020-04-28
Publication date: 2020-09-08
Also published as: WO2021217906A1

Abstract

The invention relates to artificial intelligence, and discloses a gait feature-based posture detection method, a gait feature-based posture detection device, gait feature-based posture detection equipment and a gait feature-based posture detection storage medium, wherein the method comprises the following steps: acquiring gait data of a target object, and detecting key points of a human body of the target object in the gait data to obtain a plurality of key point characteristics; performing feature classification and positioning regression on the plurality of key point features by adopting a deep learning network to obtain human body information, wherein the human body information comprises human body part positions and human body part affinity vectors; calculating matching parameters between the human body information and preset standard human body information based on the characteristics of the plurality of key points, wherein the preset standard human body information comprises a preset standard affinity vector; and determining the posture condition of the target object according to the matching parameters and a standard threshold, wherein the standard threshold is a critical value of the abnormal posture condition. In addition, the invention also relates to a block chain technology, and related data and information can be stored in the block chain nodes. The invention solves the problems of high cost and low detection efficiency when detecting the posture of the object.

Description

Gait feature-based posture detection method, device, equipment and storage medium

Technical Field

The invention relates to the field of artificial intelligence big data, in particular to a gait feature-based posture detection method, device, equipment and storage medium.

Background

Modern urban people are often troubled by cervical vertebra and lumbar vertebra diseases due to work, life and other reasons, such as disappearance of physiological curvature of cervical vertebra, protrusion of intervertebral disc and the like. These diseases are closely related to the work and study of a long-term fixed posture and are not easily perceived in the process of disease formation and development. In the early stage of disease development, medical personnel with abundant experience can distinguish and find the disease, and after the disease develops to a certain extent, medical institutions need to examine and confirm the diagnosis through medical equipment such as CT, nuclear magnetic resonance, angiography and the like.

When the target object suffers from early-stage cervical vertebra, lumbar vertebra and other diseases, the gait posture of the target object is changed due to the diseases, and when the posture change is examined and diagnosed by a medical tool, the detection cost is high, and a certain waiting period is needed to obtain a detection result, so that the detection efficiency for detecting the posture condition of the target object is low.

Disclosure of Invention

The invention mainly aims to solve the problems of high cost and low detection efficiency when detecting the posture of a target object.

The invention provides a gait feature-based posture detection method, which comprises the following steps: acquiring gait data of a target object, and detecting human key points of the target object in the gait data to obtain a plurality of key point characteristics, wherein the human key points are a plurality of coordinate points of a human skeleton; performing feature classification and positioning regression on the plurality of key point features by adopting a deep learning network to obtain human body information, wherein the human body information comprises a human body part position and a human body part affinity vector, and the human body part affinity vector is used for connecting two different key point features; calculating matching parameters between the human body information and preset standard human body information based on the plurality of key point features, wherein the preset standard human body information comprises a preset standard affinity vector, and the matching parameters are used for indicating the matching degree between the human body information of the target object and the preset standard human body information; and determining the posture condition of the target object according to the matching parameters and a standard threshold, wherein the standard threshold is a critical value of abnormal posture conditions.

Optionally, in a first implementation manner of the first aspect of the present invention, the performing, by using a deep learning network, feature classification and location regression on the multiple key point features to obtain human body information, where the human body information includes a human body part position and a human body part affinity vector, and the human body part affinity vector is used to connect two different key point features, and the implementing method includes: calculating confidence degrees between a plurality of key point features and preset human body position features by adopting a deep learning network, and classifying the key point features, wherein the preset human body position features are used for indicating different human body position, and each preset human body position feature corresponds to a human body position; when the confidence coefficient is greater than a first threshold value, determining a human body part position corresponding to the key point feature for calculating the confidence coefficient, wherein the human body part position corresponding to the key point feature is a human body part position corresponding to a preset human body position feature; calculating the affinity between two different key point features, and performing positioning regression on the two different key point features; when the affinity is greater than a second threshold value, connecting the two different key point features to generate a human body part affinity vector, wherein the human body part affinity vector is used for connecting the two different key point features; and connecting the human body part affinity vectors to obtain human body information, wherein the human body information comprises the human body part position and the human body part affinity vector.

Optionally, in a second implementation manner of the first aspect of the present invention, the calculating, based on the plurality of key point features, a matching parameter between the human body information and preset standard human body information, where the preset standard human body information includes a preset standard affinity vector, and the matching parameter is used to indicate a matching degree between the human body information of the target object and the preset standard human body information, and includes: acquiring a plurality of human body part affinity vectors of a target object; acquiring preset standard human body information matched with the target object, wherein the preset standard human body information comprises a preset standard affinity vector; calculating the confidence coefficient between the human body part affinity vector and the corresponding preset standard affinity vector by using a similarity algorithm to obtain a matching parameter, wherein the two key point characteristics in the human body part affinity vector have the same category as the preset human body position characteristics in the preset standard affinity vector, and the matching parameter is used for indicating the matching degree between the human body information of the target object and the preset standard human body information.

Optionally, in a third implementation manner of the first aspect of the present invention, the determining the posture condition of the target object according to the matching parameter and a standard threshold, where the standard threshold is a critical value of an abnormal posture condition includes: judging whether the configuration parameter is larger than a standard threshold value, wherein the standard threshold value is a critical value of an abnormal body state; and if the configuration parameter is larger than the standard threshold, determining that the posture condition of the target object is abnormal.

Optionally, in a fourth implementation manner of the first aspect of the present invention, the acquiring gait data of a target object, and detecting key points of a human body of the target object in the gait data to obtain a plurality of key point features, where the key points of the human body are a plurality of coordinate points of a human body skeleton includes: acquiring gait data of a target object, and normalizing the gait data to obtain basic processing data; and in the basic processing data, detecting human key points of the target object by adopting a convolutional neural network to obtain a plurality of key point characteristics, wherein the human key points are a plurality of coordinate points of a human skeleton.

Optionally, in a fifth implementation manner of the first aspect of the present invention, in the basic processing data, detecting a human body key point of the target object by using a convolutional neural network to obtain a plurality of key point features, where the human body key point is a plurality of coordinate points of a human body skeleton, and the detecting includes: calculating the convolution of the basic processing data by adopting a convolution neural network to obtain first processing data; performing downsampling processing on the first processing data, and extracting a plurality of sampling vectors in the first processing data to obtain second processing data; carrying out nonlinear mapping on the second processing data to obtain a gait feature map; and detecting human key points of the target object in the gait feature map to obtain a plurality of key point features, wherein the human key points are a plurality of coordinate points of a human skeleton.

Optionally, in a sixth implementation manner of the first aspect of the present invention, before the acquiring gait data of a target object, and detecting key points of a human body of the target object in the gait data to obtain a plurality of key point features, the method further includes: and calculating preset standard human body information, wherein the preset standard human body information comprises a preset standard affinity vector.

The second aspect of the present invention provides a gait feature-based posture detection device, including: the detection module is used for acquiring gait data of a target object and detecting human key points of the target object in the gait data to obtain a plurality of key point characteristics, wherein the human key points are a plurality of coordinate points of a human skeleton; the classification and regression module is used for carrying out feature classification and positioning regression on the plurality of key point features based on a deep learning network to obtain human body information, wherein the human body information comprises a human body part position and a human body part affinity vector, and the human body part affinity vector is used for connecting two different key point features; the calculation module is used for calculating matching parameters between the human body information and preset standard human body information based on the plurality of key point features, the preset standard human body information comprises a preset standard affinity vector, and the matching parameters are used for indicating the matching degree between the human body information of the target object and the preset standard human body information; and the determining module is used for determining the posture condition of the target object according to the matching parameters and a standard threshold, wherein the standard threshold is a critical value of the abnormal posture condition.

Optionally, in a first implementation manner of the second aspect of the present invention, the classification and regression module is specifically configured to: calculating confidence degrees between a plurality of key point features and preset human body position features by adopting a deep learning network, and classifying the key point features, wherein the preset human body position features are used for indicating different human body position, and each preset human body position feature corresponds to a human body position; when the confidence coefficient is greater than a first threshold value, determining a human body part position corresponding to the key point feature for calculating the confidence coefficient, wherein the human body part position corresponding to the key point feature is a human body part position corresponding to a preset human body position feature; calculating the affinity between two different key point features, and performing positioning regression on the two different key point features; when the affinity is greater than a second threshold value, connecting the two different key point features to generate a human body part affinity vector, wherein the human body part affinity vector is used for connecting the two different key point features; and connecting the human body part affinity vectors to obtain human body information, wherein the human body information comprises the human body part position and the human body part affinity vector.

Optionally, in a second implementation manner of the second aspect of the present invention, the first computing module is specifically configured to: acquiring a plurality of human body part affinity vectors of a target object; acquiring preset standard human body information matched with the target object, wherein the preset standard human body information comprises a preset standard affinity vector; calculating the confidence coefficient between the human body part affinity vector and the corresponding preset standard affinity vector by using a similarity algorithm to obtain a matching parameter, wherein the two key point characteristics in the human body part affinity vector have the same category as the preset human body position characteristics in the preset standard affinity vector, and the matching parameter is used for indicating the matching degree between the human body information of the target object and the preset standard human body information.

Optionally, in a third implementation manner of the second aspect of the present invention, the determining module is specifically configured to: judging whether the configuration parameter is larger than a standard threshold value, wherein the standard threshold value is a critical value of an abnormal body state; and if the configuration parameter is larger than the standard threshold, determining that the posture condition of the target object is abnormal.

Optionally, in a fourth implementation manner of the second aspect of the present invention, the detection module includes: the processing unit is used for acquiring gait data of a target object and normalizing the gait data to obtain basic processing data; and the detection unit is used for detecting the human key points of the target object by adopting a convolutional neural network in the basic processing data to obtain a plurality of key point characteristics, wherein the human key points are a plurality of coordinate points of a human skeleton.

Optionally, in a fifth implementation manner of the second aspect of the present invention, the detecting unit is specifically configured to calculate a convolution of the basic processing data by using a convolutional neural network, so as to obtain first processing data; performing downsampling processing on the first processing data, and extracting a plurality of sampling vectors in the first processing data to obtain second processing data; carrying out nonlinear mapping on the second processing data to obtain a gait feature map; and detecting human key points of the target object in the gait feature map to obtain a plurality of key point features, wherein the human key points are a plurality of coordinate points of a human skeleton.

Optionally, in a sixth implementation manner of the second aspect of the present invention, the gait feature-based posture detecting apparatus further includes: and the second calculation module is used for calculating preset standard human body information, and the preset standard human body information comprises a preset standard affinity vector.

A third aspect of the present invention provides a gait feature-based posture detection apparatus, including: a memory having instructions stored therein and at least one processor, the memory and the at least one processor interconnected by a line; the at least one processor invokes the instructions in the memory to cause the gait feature based posture detection apparatus to perform the gait feature based posture detection method described above.

A fourth aspect of the present invention provides a computer-readable storage medium including a storage data area storing data created according to use of a blockchain node and a storage program area storing a computer program, wherein the computer program, when executed by a processor, implements the gait feature-based posture detection method described above.

According to the technical scheme provided by the invention, gait data of a target object are obtained, and human key points of the target object are detected in the gait data to obtain a plurality of key point characteristics, wherein the human key points are a plurality of coordinate points of a human skeleton; performing feature classification and positioning regression on the plurality of key point features by adopting a deep learning network to obtain human body information, wherein the human body information comprises a human body part position and a human body part affinity vector, and the human body part affinity vector is used for connecting two different key point features; calculating matching parameters between the human body information and preset standard human body information based on the plurality of key point features, wherein the preset standard human body information comprises a preset standard affinity vector, and the matching parameters are used for indicating the matching degree between the human body information of the target object and the preset standard human body information; and determining the posture condition of the target object according to the matching parameters and a standard threshold, wherein the standard threshold is a critical value of abnormal posture conditions. In the embodiment of the invention, the gait data of the target object is processed and analyzed by the convolutional neural network and the deep learning network, the matching parameters between the human body information obtained after processing and the preset standard human body information are calculated, and the posture state of the target object is determined by the matching parameters. The gait data of the target object is processed through the convolutional neural network and the deep learning network, so that the detection cost is reduced, and the detection efficiency of detecting the body state condition of the target object is improved.

Drawings

FIG. 1 is a schematic diagram of an embodiment of a gait feature-based posture detection method according to an embodiment of the invention;

FIG. 2 is a schematic diagram of another embodiment of a gait feature-based posture detection method according to an embodiment of the invention;

FIG. 3 is a schematic diagram of an embodiment of a gait feature-based posture detection device according to an embodiment of the invention;

FIG. 4 is a schematic diagram of another embodiment of a gait feature-based posture detection device in an embodiment of the invention;

fig. 5 is a schematic diagram of an embodiment of a gait feature-based posture detection device in the embodiment of the invention.

Detailed Description

The embodiment of the invention provides a gait feature-based posture detection method, a gait feature-based posture detection device, gait feature-based posture detection equipment and a storage medium. The gait data of the target object is processed through the convolutional neural network and the deep learning network, so that the detection cost is reduced, and the detection efficiency of detecting the body state condition of the target object is improved.

The terms "first," "second," "third," "fourth," and the like in the description and in the claims, as well as in the drawings, if any, are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It will be appreciated that the data so used may be interchanged under appropriate circumstances such that the embodiments described herein may be practiced otherwise than as specifically illustrated or described herein. Furthermore, the terms "comprises," "comprising," or "having," and any variations thereof, are intended to cover non-exclusive inclusions, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed, but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.

For convenience of understanding, a detailed flow of an embodiment of the present invention is described below, and referring to fig. 1, an embodiment of a gait feature-based posture detection method according to an embodiment of the present invention includes:

101. acquiring gait data of a target object, and detecting human key points of the target object in the gait data to obtain a plurality of key point characteristics, wherein the human key points are a plurality of coordinate points of a human skeleton;

it is understood that the executing subject of the present invention may be a gait feature-based posture detection device, and may also be a terminal or a server, which is not limited herein. The embodiment of the present invention is described by taking a server as an execution subject.

The server acquires gait data of a target object, processes the gait data, and then detects human key points of the target object in the processed gait data to further obtain a plurality of key point characteristics, wherein the human key points are a plurality of coordinate points in a human skeleton of the target object, and the coordinate points are indispensable representative points in the human skeleton.

The gait data of the target object acquired by the server refers to gait video data of the target object walking continuously, and the posture state of the target object is determined by analyzing the human body information of the target object in the video data. The server detects human body key points of a target object in gait data through a Convolutional Neural Network (CNN) and obtains a plurality of key feature points, wherein the CNN is a type of feedforward neural network which contains convolution calculation and has a deep structure, and is one of the representative algorithms of deep learning. The CNN has the characteristic learning ability, can carry out translation invariant classification on input information according to the hierarchical structure of the CNN, and can realize the processing of gait data according to the ability of the CNN so as to obtain a gait feature map, so that the server can detect key points of a human body in the gait feature map.

102. Performing feature classification and positioning regression on the plurality of key point features by adopting a deep learning network to obtain human body information, wherein the human body information comprises a human body part position and a human body part affinity vector, and the human body part affinity vector is used for connecting two different key point features;

the server performs feature classification and positioning regression on the plurality of key point features by adopting a deep learning network, determines the category of the key point features, and obtains human body information of the target object, wherein the human body information comprises a plurality of human body part positions and human body part affinity vectors, and the human body part affinity vectors are used for connecting two different key point features.

The server adopts a human body key point detection algorithm from bottom to top, mainly comprises two parts, namely key point classification and key point positioning regression, wherein the key point detection requires the server to calculate the confidence degrees between all key point features and the preset human body position features, to judge the category of the key point characteristics and determine the corresponding human body position represented by the key point characteristics, then the server calculates the affinity between different key point characteristics, connects the different key point characteristics together according to the size of the affinity, the human body part affinity vector is determined, the human body part affinity vector is the basis for forming a human body skeleton, a plurality of human body part affinity vectors are connected together to form the human body skeleton, the human skeleton here represents the approximate form of human motion, and the server obtains the detected human body information from the human body part position and the human body part affinity vector.

After the server determines the classification of all the key point features, the key point features need to be positioned and regressed, that is, different key point features are connected together with high affinity to obtain human body part affinity vectors. The affinity here refers to the degree of association between two different key point features, and when the degree of association between two different key point features is greater than a second threshold, it indicates that the greater the degree of association between the two different key point features, the two different key point features are connected together to form a human body part affinity vector, and a plurality of human body part affinity vectors are connected together to form a human body skeleton. One keypoint feature may connect a different number of other keypoint features, such as: the hand keypoint features are connected only to the elbow keypoint features, and the shoulder keypoint features are connected to the elbow keypoint features and the torso keypoint features, respectively.

103. Calculating matching parameters between the human body information and preset standard human body information based on the plurality of key point features, wherein the preset standard human body information comprises a preset standard affinity vector, and the matching parameters are used for indicating the matching degree between the human body information of the target object and the preset standard human body information;

the server calculates matching parameters between the human body information and preset standard human body information according to the relation between the plurality of key point features, the preset standard human body information comprises a preset standard affinity vector, the preset standard affinity vector is obtained by inputting a large amount of data into the model for calculation, and the matching parameters are used for indicating the matching degree between the human body information of the target object and the preset standard human body information.

After the server calculates the human body part affinity vectors of the plurality of key point features, the human body part affinity vectors are connected to obtain a human body posture skeleton, the human body posture skeleton is compared and analyzed with the standard posture skeleton, and the server can obtain the posture condition of the target object. The human posture skeleton is compared with a standard posture skeleton, and is essentially the comparison between human body part affinity vectors at a plurality of same positions and preset standard affinity vectors, wherein the human body part affinity vectors are the basis for forming the human posture skeleton, the preset standard affinity vectors are the basis for forming the standard posture skeleton, the preset standard affinity vectors are obtained by continuously calculating and training a large number of human posture samples, and the human postures with different heights and weights have different standard posture skeletons and preset standard affinity vectors.

104. And determining the posture condition of the target object according to the matching parameters and a standard threshold, wherein the standard threshold is a critical value of the abnormal posture condition.

And after the server obtains the matching parameters by calculating the confidence coefficient between the human body part affinity vector and the preset standard affinity vector, comparing the matching parameters with the standard threshold values of different posture conditions, and further determining the posture condition of the target object.

The standard threshold of the posture condition is used for explaining the critical value of different posture conditions, and the standard threshold is calculated by a large amount of data. Such as: the standard threshold value of the cervical vertebra abnormality is 0.8, that is, when the matching parameter is greater than the standard threshold value of 0.8, the posture condition of the target object is the cervical vertebra abnormality. In addition, the standard threshold is at least one, each standard threshold represents a critical value of different posture conditions, the standard threshold and the matching parameter are corresponding to each other, that is, the human body part affinity vector for calculating the matching parameter is related to the posture condition represented by the standard threshold. For example: the standard threshold value of the cervical vertebra abnormality is 0.8, and the key point features of the human body part affinity vector forming the calculation matching parameters are related to the neck and can be head key point features, shoulder key point features and trunk key point features. And calculating the matching parameters for multiple times by using different human body part affinity vectors to obtain more accurate posture conditions of the target object.

It is understood that the posture status of the target object obtained by the server can be various, such as: the standard threshold value of the slight cervical vertebra abnormality is 0.65, namely, when the matching parameter is greater than the standard threshold value of 0.65, the posture condition of the target object is the slight cervical vertebra abnormality; such as: the standard threshold value of the serious cervical vertebra abnormality is 0.89, that is, when the matching parameter is greater than the standard threshold value of 0.89, the posture condition of the target subject is the serious cervical vertebra abnormality. Different human body part affinity vectors can be obtained through different key point characteristics, and the abnormal posture conditions of the target object compared with the standard health conditions can be obtained by comparing the different human body part affinity vectors with the preset standard affinity vectors through the server.

In the embodiment of the invention, the gait data of the target object is processed and analyzed by adopting an artificial intelligence convolutional neural network and a deep learning network, the matching parameters between the human body information obtained after processing and the preset standard human body information are calculated, and the posture condition of the target object is determined through the matching parameters. Meanwhile, the invention can also be applied to the field of intelligent medical treatment, thereby promoting the construction of a smart city, and the invention processes the gait data of the target object through the convolutional neural network and the deep learning network, reduces the detection cost and improves the detection efficiency of detecting the posture condition of the target object

Referring to fig. 2, another embodiment of the gait feature-based posture detection method according to the embodiment of the invention includes:

201. calculating preset standard human body information, wherein the preset standard human body information comprises a preset standard affinity vector;

the server adopts a deep learning network and a convolutional neural network to calculate preset standard human body information, wherein the preset standard human body information comprises a preset standard affinity vector.

It can be understood that the preset standard human body information includes human body information under standard health condition and human body information suffering from cervical and lumbar diseases of different degrees, the preset standard human body information is obtained by collecting a large amount of human body information data by the server, performing deep learning training on the large amount of human body information data, extracting feature extraction layers of gait data with different abstract degree features through a deep learning network, performing feature classification and positioning regression on key point features of the human body, evaluating the obtained training result by using a relevant loss function, updating network parameters by back propagation, and repeating the process until the network converges to obtain the preset standard human body information.

It should be noted that, the method for calculating the preset standard human body information by the server is the same as the method for calculating the human body information, and both the method for calculating the preset standard human body information and the method for calculating the human body information utilize the convolutional neural network to process the gait data, and then utilize the deep learning network to perform feature classification and positioning regression on the human body key points, so as to obtain the human body information, and thus the obtained data result is more accurate and more representative.

202. Acquiring gait data of a target object, and detecting human key points of the target object in the gait data to obtain a plurality of key point characteristics, wherein the human key points are a plurality of coordinate points of a human skeleton;

the server acquires gait data of a target object, processes the gait data, and then detects human key points of the target object in the processed gait data to further obtain a plurality of key point characteristics, wherein the human key points are a plurality of coordinate points in a human skeleton of the target object, and the coordinate points are indispensable representative points in the human skeleton. Specifically, the method comprises the following steps:

the server carries out normalization processing on the gait data of the target object to obtain basic processing data; in the basic processing data, the server detects human key points of a target object by adopting a convolutional neural network to obtain a plurality of key point characteristics, wherein the human key points are a plurality of coordinate points of a human skeleton. Specifically, the method comprises the following steps: firstly, a server calculates the convolution of basic processing data by adopting a convolution neural network to obtain first processing data; secondly, the server performs downsampling processing on the first processing data, extracts a plurality of sampling vectors in the first processing data, and obtains second processing data; then the server carries out nonlinear mapping on the second processing data to obtain a gait feature map; and finally, the server detects human key points of the target object in the gait feature map to obtain a plurality of key point features, wherein the human key points are a plurality of coordinate points of the human skeleton.

The gait data of the target object acquired by the server refers to gait video data of the target object walking continuously, and the posture state of the target object is determined by analyzing the human body information of the target object in the video data. It can be understood that before the server detects the human body key points of the target object through the gait data, the server needs to normalize the gait data of the target object, and the server normalizes the video data, which is beneficial to the next step. The server acquires pixel values on the gait data, and the server normalizes the pixel values to acquire basic processing data. The normalization does not change the contrast of the image, and simultaneously ensures that all the pixel values of the image after normalization are in the range of [0, 1 ]. The formula used is as follows:

in the formula: a' is the pixel value of the basic processing data, and a is the original pixel value of the gait data. It can be understood that the gait data of the target object is subjected to normalization processing, the data are mapped to a range of 0-1 for processing, the obtained basic processing data are input into the network model, and the server is more convenient and faster to calculate.

After obtaining the basic processing data, the server detects human key points of the target object in the gait data through a Convolutional Neural Network (CNN) and obtains a plurality of key feature points, wherein the CNN is a feedforward neural network which contains convolutional calculation and has a deep structure and is one of the representative algorithms of deep learning. The CNN has the characteristic learning ability, can carry out translation invariant classification on input information according to the hierarchical structure of the CNN, and can realize the processing of gait data according to the ability of the CNN so as to obtain a gait feature map, so that the server can detect key points of a human body in the gait feature map. The CNN includes a feature extractor composed of convolutional layers and sub-sampling layers, and extracts features of different levels by repeatedly processing the basic processing data by using the feature extractor to operate different parameters.

In the convolutional layer of the convolutional neural network, a server calculates the convolution of basic processing data, and in a convolutional layer of the CNN, a plurality of feature planes are usually included, each feature plane is composed of a plurality of neurons arranged in a rectangle, the neurons of the same feature plane share a weight, and the shared weight is a convolutional kernel. The convolution kernel is generally initialized in the form of a random decimal matrix, and the convolution kernel learns to obtain a reasonable weight in the training process of the network, namely the server calculates the convolution of basic processing data to further obtain first processing data; after obtaining the first processed data, the server performs downsampling processing on the first processed data, the downsampling is also called pooling, and can be regarded as a special convolution process for reducing the size of the feature map, and after the downsampling processing is performed, a plurality of sampling vectors are extracted to obtain second processed data; secondly, the server adopts an excitation function to carry out nonlinear mapping on the second processing data so as to obtain a gait feature map, and feature points in different forms can be displayed in the gait feature map; and finally, the server screens out human key points of the target object from the gait feature map to obtain a plurality of key point features, wherein the human key points are coordinate points specifically representing human skeleton combinations, and human body information of the target object can be obtained through the key feature points, so that the posture condition of the target object is better determined.

203. Performing feature classification and positioning regression on the plurality of key point features by adopting a deep learning network to obtain human body information, wherein the human body information comprises a human body part position and a human body part affinity vector, and the human body part affinity vector is used for connecting two different key point features;

the server performs feature classification and positioning regression on the plurality of key point features by adopting a deep learning network, determines the category of the key point features, and obtains human body information of the target object, wherein the human body information comprises a plurality of human body part positions and human body part affinity vectors, and the human body part affinity vectors are used for connecting two different key point features. Specifically, the method comprises the following steps:

the server firstly adopts a deep learning network to calculate confidence degrees between a plurality of key point characteristics and preset human body position characteristics so as to obtain the categories of the key point characteristics, wherein the preset human body position characteristics are used for indicating different human body position, and each preset human body position characteristic corresponds to a human body position; secondly, the server compares the confidence with a first threshold, and when the confidence is greater than the first threshold, the position of the human body part corresponding to the key point feature of the calculated confidence is determined, wherein the position of the human body part corresponding to the key point feature is the position of the human body part corresponding to the preset human body position feature; then the server calculates the affinity between two different key point characteristics and performs positioning regression on the two different key point characteristics; the server compares the affinity with a second threshold, and when the affinity is greater than the second threshold, connects two different key point features, wherein the corresponding two different key point features are the two different key point features for calculating the affinity, and generates a human body part affinity vector, wherein the human body part affinity vector is used for connecting the two different key point features; and finally, connecting the plurality of human body part affinity vectors by the server to obtain human body information, wherein the human body information comprises the human body part position and the human body part affinity vectors.

It should be noted that the preset human body position features herein refer to standard human body bone key points, and there are many preset human body position features, such as: the method comprises the steps that 18 standard heads (5), shoulders (2), trunks (1), elbows (2), hands (2), hips (2), knees (2) and feet (2) of human skeletons are arranged, a server determines classification of key point features by calculating confidence degrees between preset human body position features and recognized key point features, when the confidence degrees between the key point features and the preset human body position features are larger than a first threshold value, the key point features are matched with the preset human body position features, and the server sets the classification of the key point features as classification of corresponding preset human body position features. The value of the first threshold may be set differently depending on the actual situation, and is not limited in the present application.

For example, the server sets the first threshold to be 0.6, the server calculates the confidence coefficient between the key point feature and the preset elbow feature to be 0.3, and then calculates the confidence coefficient between the key point feature and the preset knee feature to be 0.8, and the confidence coefficient between the key point feature and other preset body position features does not exceed 0.8, which indicates that the key point feature is matched with the preset knee feature, and the server determines that the key point feature is the preset knee feature.

For example, the server sets the second threshold to be 0.55, the server calculates that the affinity between the trunk key point feature and the shoulder key point feature is 0.88, the affinity between the trunk key point feature and the foot key point feature is 0.15, and the affinity between the trunk key point feature and the head key point feature is 0.75, and when the calculated affinity is greater than the second preset 0.55, the server connects the two key point features for calculating the affinity together, that is, connects the trunk key point feature with the shoulder key point feature and the head key point feature respectively, so as to obtain the human body part affinity vector.

204. Calculating matching parameters between the human body information and preset standard human body information based on the plurality of key point features, wherein the preset standard human body information comprises a preset standard affinity vector, and the matching parameters are used for indicating the matching degree between the human body information of the target object and the preset standard human body information;

the server calculates matching parameters between the human body information and preset standard human body information according to the relation between the plurality of key point features, the preset standard human body information comprises a preset standard affinity vector, the preset standard affinity vector is obtained by inputting a large amount of data into the model for calculation, and the matching parameters are used for indicating the matching degree between the human body information of the target object and the preset standard human body information. Specifically, the method comprises the following steps:

the server firstly obtains a plurality of human body part affinity vectors of a target object; then the server acquires preset standard human body information matched with the target object, wherein the preset standard human body information comprises a plurality of different preset standard affinity vectors; and finally, the server calculates the confidence coefficient between the human body part affinity vector and the corresponding preset standard affinity vector by using a similarity algorithm to obtain a matching parameter, wherein the two key point characteristics in the human body part affinity vector are the same as the preset human body position characteristics in the preset standard affinity vector in category, and the matching parameter is used for indicating the matching degree between the human body information of the target object and the preset standard human body information.

After the server calculates the human body part affinity vectors of the plurality of key point features, the human body part affinity vectors are connected to obtain a human body posture skeleton, the human body posture skeleton is compared and analyzed with the standard posture skeleton, and the server can obtain the posture condition of the target object. The human posture skeleton is compared with a standard posture skeleton, the comparison between human body part affinity vectors at a plurality of same positions and preset standard affinity vectors is essentially carried out, the human body part affinity vectors are the basis for forming the human posture skeleton, the preset standard affinity vectors are the basis for forming the standard posture skeleton, the preset standard affinity vectors are obtained by continuously calculating and training a large number of human posture samples, human postures with different heights and weights have different standard posture skeletons and preset standard affinity vectors, and therefore when the matching parameters are compared with the standard threshold values, the preset standard human body information for calculating the matching parameters is necessary to be matched with a target object.

It should be noted that, when the confidence between the human body part affinity vector and the preset standard affinity vector is calculated by using the similarity calculation method, the two key point features in the human body part affinity vector are the same as the preset human body position features in the preset standard affinity vector in category, that is, the two key feature points forming the human body part affinity vector are the same as the two preset human body position features forming the preset standard affinity vector in category.

For example, taking the comparison of human body part affinity vectors between the head key point features and the trunk key point features of the target object as an example, firstly, the server obtains the human body part affinity vectors between the head key point features and the trunk key point features, which are hereinafter referred to as first affinity vectors, secondly, the server screens out the preset standard human body information matched with the target object, where the preset standard human body information matched with the target object refers to the human posture skeleton information based on the standard health of the target object, obtains the preset standard affinity vectors between the standard head key point features and the standard trunk key point features, which are hereinafter referred to as second affinity vectors, and finally, the server calculates the confidence degree between the first affinity vectors and the second affinity vectors, that is, calculates the matching degree between the first affinity vectors and the second affinity vectors, a matching parameter is obtained.

205. And determining the posture condition of the target object according to the matching parameters and a standard threshold, wherein the standard threshold is a critical value of the abnormal posture condition.

And after the server obtains the matching parameters by calculating the confidence coefficient between the human body part affinity vector and the preset standard affinity vector, comparing the matching parameters with the standard threshold values of different posture conditions, and further determining the posture condition of the target object. Specifically, the method comprises the following steps: the server judges whether the configuration parameters are larger than a standard threshold value, wherein the standard threshold value is a critical value of the abnormal body state; and if the configuration parameter is larger than the standard threshold, the server determines that the posture condition of the target object is abnormal.

In the embodiment of the invention, the gait data of the target object is processed and analyzed by the convolutional neural network and the deep learning network, the matching parameters between the human body information obtained after processing and the preset standard human body information are calculated, and the posture state of the target object is determined by the matching parameters. The gait data of the target object is processed through the convolutional neural network and the deep learning network, so that the detection cost is reduced, and the detection efficiency of detecting the body state condition of the target object is improved.

With reference to fig. 3, the method for detecting a posture based on gait features in the embodiment of the present invention is described above, and a device for detecting a posture based on gait features in the embodiment of the present invention is described below, where an embodiment of the device for detecting a posture based on gait features in the embodiment of the present invention includes:

the detection module 301 is configured to acquire gait data of a target object, and detect key points of a human body of the target object in the gait data to obtain a plurality of key point features, where the key points of the human body are a plurality of coordinate points of a human body skeleton;

a classification and regression module 302, configured to perform feature classification and positioning regression on the multiple key point features based on a deep learning network to obtain human body information, where the human body information includes a human body part position and a human body part affinity vector, and the human body part affinity vector is used to connect two different key point features;

a first calculating module 303, configured to calculate matching parameters between the human body information and preset standard human body information based on the multiple key point features, where the preset standard human body information includes a preset standard affinity vector, and the matching parameters are used to indicate a matching degree between the human body information of the target object and the preset standard human body information;

the determining module 304 is configured to determine the posture status of the target object according to the matching parameter and a standard threshold, where the standard threshold is a critical value of the abnormal posture status.

Referring to fig. 4, another embodiment of the posture detection device based on gait characteristics according to the embodiment of the present invention includes:

Optionally, the classification and regression module 302 may be further specifically configured to:

calculating confidence degrees between the plurality of key point features and preset human body position features by adopting a deep learning network, classifying the plurality of key point features, wherein the preset human body position features are used for indicating different human body position, and each preset human body position feature corresponds to a human body position;

when the confidence coefficient is greater than a first threshold value, determining the human body part position corresponding to the key point feature for calculating the confidence coefficient, wherein the human body part position corresponding to the key point feature is the human body part position corresponding to the preset human body position feature;

calculating the affinity between two different key point characteristics, and performing positioning regression on the two different key point characteristics;

when the affinity is greater than a second threshold value, connecting two different key point features to generate a human body part affinity vector, wherein the human body part affinity vector is used for connecting the two different key point features;

and connecting the plurality of human body part affinity vectors to obtain human body information, wherein the human body information comprises the human body part position and the human body part affinity vector.

Optionally, the first calculating module 303 may be further specifically configured to:

acquiring a plurality of human body part affinity vectors of a target object;

acquiring preset standard human body information matched with a target object, wherein the preset standard human body information comprises a preset standard affinity vector;

and calculating the confidence coefficient between the human body part affinity vector and the corresponding preset standard affinity vector by using a similarity algorithm to obtain a matching parameter, wherein the two key point characteristics in the human body part affinity vector are the same as the preset human body position characteristics in the preset standard affinity vector in category, and the matching parameter is used for indicating the matching degree between the human body information of the target object and the preset standard human body information.

Optionally, the determining module 304 may be further specifically configured to:

judging whether the configuration parameters are larger than a standard threshold value, wherein the standard threshold value is a critical value of the abnormal body state;

and if the configuration parameter is larger than the standard threshold, determining that the posture condition of the target object is abnormal.

Optionally, the detection module 301 includes:

the processing unit 3011 is configured to obtain gait data of the target object, and normalize the gait data to obtain basic processing data;

the detecting unit 3012 is configured to detect, in the basic processing data, a human body key point of the target object by using a convolutional neural network, to obtain a plurality of key point features, where the human body key point is a plurality of coordinate points of a human body skeleton.

Optionally, the detecting unit 3012 may be further specifically configured to:

calculating the convolution of the basic processing data by adopting a convolution neural network to obtain first processing data;

performing downsampling processing on the first processing data, and extracting a plurality of sampling vectors in the first processing data to obtain second processing data;

carrying out nonlinear mapping on the second processing data to obtain a gait feature map;

and detecting human key points of the target object in the gait feature map to obtain a plurality of key point features, wherein the human key points are a plurality of coordinate points of a human skeleton.

Optionally, the posture detecting device based on gait characteristics further includes:

and a second calculating module 305, configured to calculate preset standard human body information, where the preset standard human body information includes a preset standard affinity vector.

The gait feature-based posture detection device in the embodiment of the invention is described in detail in terms of the modular functional entity in fig. 3 and 4, and the gait feature-based posture detection device in the embodiment of the invention is described in detail in terms of the hardware processing.

Fig. 5 is a schematic structural diagram of a gait feature-based posture detection apparatus 500 according to an embodiment of the present invention, which may generate relatively large differences due to different configurations or performances, and may include one or more processors (CPUs) 510 (e.g., one or more processors) and a memory 520, one or more storage media 530 (e.g., one or more mass storage devices) storing an application 533 or data 532. Memory 520 and storage media 530 may be, among other things, transient or persistent storage. The program stored on the storage medium 530 may include one or more modules (not shown), each of which may include a series of instructions operating on the gait feature based posture detection apparatus 500. Still further, the processor 510 may be configured to communicate with the storage medium 530 to execute a series of instruction operations in the storage medium 530 on the gait feature based posture detecting apparatus 500.

The gait feature-based posture sensing apparatus 500 may also include one or more power supplies 540, one or more wired or wireless network interfaces 550, one or more input-output interfaces 560, and/or one or more operating systems 531, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art will appreciate that the gait feature based posture detection device configuration shown in fig. 5 does not constitute a limitation of the gait feature based posture detection device and may include more or fewer components than those shown, or some components in combination, or a different arrangement of components.

The present invention also provides a computer readable storage medium, which may be a non-volatile computer readable storage medium, which may also be a volatile computer readable storage medium, having stored therein instructions, which, when executed on a computer, cause the computer to perform the steps of the gait feature based posture detection method.

Further, the computer-readable storage medium may mainly include a storage program area and a storage data area, wherein the storage program area may store an operating system, an application program required for at least one function, and the like; the storage data area may store data created according to the use of the blockchain node, and the like.

It is clear to those skilled in the art that, for convenience and brevity of description, the specific working processes of the above-described systems, apparatuses and units may refer to the corresponding processes in the foregoing method embodiments, and are not described herein again.

The integrated unit, if implemented in the form of a software functional unit and sold or used as a stand-alone product, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present invention may be embodied in the form of a software product, which is stored in a storage medium and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: various media capable of storing program codes, such as a usb disk, a removable hard disk, a read-only memory (ROM), a Random Access Memory (RAM), a magnetic disk, or an optical disk.

The block chain is a novel application mode of computer technologies such as distributed data storage, point-to-point transmission, a consensus mechanism, an encryption algorithm and the like. A block chain (Blockchain), which is essentially a decentralized database, is a series of data blocks associated by using a cryptographic method, and each data block contains information of a batch of network transactions, so as to verify the validity (anti-counterfeiting) of the information and generate a next block. The blockchain may include a blockchain underlying platform, a platform product service layer, an application service layer, and the like.

The above embodiments are only used to illustrate the technical solution of the present invention, and not to limit the same; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.

Claims

1. A gait feature-based posture detection method is characterized by comprising the following steps:

acquiring gait data of a target object, and detecting human key points of the target object in the gait data to obtain a plurality of key point characteristics, wherein the human key points are a plurality of coordinate points of a human skeleton;

performing feature classification and positioning regression on the plurality of key point features by adopting a deep learning network to obtain human body information, wherein the human body information comprises a human body part position and a human body part affinity vector, and the human body part affinity vector is used for connecting two different key point features;

calculating matching parameters between the human body information and preset standard human body information based on the plurality of key point features, wherein the preset standard human body information comprises a preset standard affinity vector, and the matching parameters are used for indicating the matching degree between the human body information of the target object and the preset standard human body information;

and determining the posture condition of the target object according to the matching parameters and a standard threshold, wherein the standard threshold is a critical value of abnormal posture conditions.

2. The gait feature-based posture detection method according to claim 1, wherein the performing feature classification and localization regression on the plurality of key point features by using a deep learning network to obtain human body information, the human body information including a human body part position and a human body part affinity vector, the human body part affinity vector for connecting two different key point features comprises:

calculating confidence degrees between a plurality of key point features and preset human body position features by adopting a deep learning network, and classifying the key point features, wherein the preset human body position features are used for indicating different human body position, and each preset human body position feature corresponds to a human body position;

when the confidence coefficient is greater than a first threshold value, determining a human body part position corresponding to the key point feature for calculating the confidence coefficient, wherein the human body part position corresponding to the key point feature is a human body part position corresponding to a preset human body position feature;

calculating the affinity between two different key point features, and performing positioning regression on the two different key point features;

when the affinity is greater than a second threshold value, connecting the two different key point features to generate a human body part affinity vector, wherein the human body part affinity vector is used for connecting the two different key point features;

and connecting the human body part affinity vectors to obtain human body information, wherein the human body information comprises the human body part position and the human body part affinity vector.

3. The gait feature-based posture detection method according to claim 2, wherein the calculating a matching parameter between the body information and preset standard body information based on the plurality of key point features, the preset standard body information including a preset standard affinity vector, the matching parameter indicating a degree of matching between the body information of the target subject and the preset standard body information includes:

acquiring a human body part affinity vector of a target object;

acquiring preset standard human body information matched with the target object, wherein the preset standard human body information comprises a plurality of preset standard affinity vectors;

calculating the confidence coefficient between the human body part affinity vector and the corresponding preset standard affinity vector by using a similarity algorithm to obtain a matching parameter, wherein the two key point characteristics in the human body part affinity vector have the same category as the preset human body position characteristics in the preset standard affinity vector, and the matching parameter is used for indicating the matching degree between the human body information of the target object and the preset standard human body information.

4. The gait feature-based posture detection method according to claim 1, wherein the determining the posture condition of the target subject according to the matching parameter and a standard threshold, the standard threshold being a threshold of abnormal posture conditions comprises:

judging whether the configuration parameter is larger than a standard threshold value, wherein the standard threshold value is a critical value of an abnormal body state;

5. The gait feature-based posture detection method according to claim 1, wherein the acquiring gait data of a target object and detecting key points of a human body of the target object in the gait data to obtain a plurality of key point features, the key points of the human body being a plurality of coordinate points of a human body skeleton comprises:

acquiring gait data of a target object, and normalizing the gait data to obtain basic processing data;

and in the basic processing data, detecting human key points of the target object by adopting a convolutional neural network to obtain a plurality of key point characteristics, wherein the human key points are a plurality of coordinate points of a human skeleton.

6. The gait feature-based posture detection method according to claim 5, wherein the detecting a plurality of key point features of a human body key point of the target object by using a convolutional neural network in the basic processing data, wherein the detecting the human body key point is a plurality of coordinate points of a human body skeleton comprises:

7. A gait feature-based posture detection method according to any one of claims 1-6, characterized in that before acquiring gait data of a target object and detecting key points of a human body of the target object in the gait data to obtain a plurality of key point features, the method further comprises:

and calculating preset standard human body information, wherein the preset standard human body information comprises a preset standard affinity vector.

8. A gait feature-based posture detection device, comprising:

the detection module is used for acquiring gait data of a target object and detecting human key points of the target object in the gait data to obtain a plurality of key point characteristics, wherein the human key points are a plurality of coordinate points of a human skeleton;

the classification and regression module is used for carrying out feature classification and positioning regression on the plurality of key point features based on a deep learning network to obtain human body information, wherein the human body information comprises a human body part position and a human body part affinity vector, and the human body part affinity vector is used for connecting two different key point features;

the calculation module is used for calculating matching parameters between the human body information and preset standard human body information based on the plurality of key point features, the preset standard human body information comprises a preset standard affinity vector, and the matching parameters are used for indicating the matching degree between the human body information of the target object and the preset standard human body information;

and the determining module is used for determining the posture condition of the target object according to the matching parameters and a standard threshold, wherein the standard threshold is a critical value of the abnormal posture condition.

9. A gait feature-based posture detection apparatus characterized by comprising: a memory having instructions stored therein and at least one processor, the memory and the at least one processor interconnected by a line;

the at least one processor invokes the instructions in the memory to cause the gait feature based posture detection apparatus to perform a gait feature based posture detection method of any one of claims 1-7.

10. A computer-readable storage medium comprising a data storage area storing data created according to use of blockchain nodes and a program storage area storing a computer program, wherein the computer program, when executed by a processor, implements the gait feature based posture detection method according to any one of claims 1 to 7.