CN116226727A - Motion recognition system based on AI - Google Patents
Motion recognition system based on AI Download PDFInfo
- Publication number
- CN116226727A CN116226727A CN202310491790.7A CN202310491790A CN116226727A CN 116226727 A CN116226727 A CN 116226727A CN 202310491790 A CN202310491790 A CN 202310491790A CN 116226727 A CN116226727 A CN 116226727A
- Authority
- CN
- China
- Prior art keywords
- data
- module
- angle
- human body
- acquisition
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 230000033001 locomotion Effects 0.000 title claims abstract description 75
- 238000012545 processing Methods 0.000 claims abstract description 12
- 238000000605 extraction Methods 0.000 claims abstract description 8
- 238000004458 analytical method Methods 0.000 claims abstract description 7
- 238000000034 method Methods 0.000 claims description 19
- 239000011159 matrix material Substances 0.000 claims description 16
- 230000009471 action Effects 0.000 claims description 14
- 239000000284 extract Substances 0.000 claims description 10
- 230000002159 abnormal effect Effects 0.000 claims description 9
- 239000013598 vector Substances 0.000 claims description 9
- 230000001133 acceleration Effects 0.000 claims description 8
- 230000008569 process Effects 0.000 claims description 6
- 238000007781 pre-processing Methods 0.000 claims description 5
- 238000006243 chemical reaction Methods 0.000 claims description 4
- 238000005070 sampling Methods 0.000 claims description 4
- 238000012360 testing method Methods 0.000 claims description 4
- 238000013528 artificial neural network Methods 0.000 claims description 3
- 238000004364 calculation method Methods 0.000 claims description 3
- 238000012512 characterization method Methods 0.000 claims description 3
- 239000000470 constituent Substances 0.000 claims description 3
- 238000010276 construction Methods 0.000 claims description 3
- 238000010183 spectrum analysis Methods 0.000 claims description 3
- 238000007619 statistical method Methods 0.000 claims description 3
- 238000004422 calculation algorithm Methods 0.000 claims description 2
- 238000013527 convolutional neural network Methods 0.000 claims description 2
- 238000012937 correction Methods 0.000 claims description 2
- 238000012549 training Methods 0.000 claims description 2
- 230000036544 posture Effects 0.000 description 6
- 230000007547 defect Effects 0.000 description 3
- 238000012544 monitoring process Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000004044 response Effects 0.000 description 2
- 230000000007 visual effect Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000006073 displacement reaction Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 239000011664 nicotinic acid Substances 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000005096 rolling process Methods 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Images
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/103—Detecting, measuring or recording devices for testing the shape, pattern, colour, size or movement of the body or parts thereof, for diagnostic purposes
- A61B5/11—Measuring movement of the entire body or parts thereof, e.g. head or hand tremor, mobility of a limb
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D30/00—Reducing energy consumption in communication networks
- Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks
Abstract
The invention discloses a motion recognition system based on AI, which relates to the technical field of motion recognition and comprises a sensor module, a receiving module, an acquisition module, an extraction module, a processing module and a recognition module, wherein in order to improve the accuracy of motion analysis, human body motions are limited according to the actual condition of human body motions, not all limb space changes are human body motions, the motions have periodicity or obvious changes, on the basis, the AI can automatically screen limb motions without periodicity or obvious changes, the data volume of acquired data is further reduced, the problem of slow judgment caused by overlarge acquired data of the system can be reasonably avoided, capturing equipment needs to acquire at the frequency of 100HZ, and the frequency can not take excessive data in a short time under the premise of ensuring the adoption accuracy.
Description
Technical Field
The invention relates to the technical field of motion recognition, in particular to an AI-based motion recognition system.
Background
The human body movement is completed under the control of a related system of a person, is complex, various and delicate, has flexibility and variability which cannot be achieved by any machine, and has important pushing effects on bionic engineering, medical engineering, sports competition, game animation and the like.
The traditional human motion research is realized by analyzing video or image sequences, firstly detecting a moving human body, extracting the general outline of the moving human body, then tracking the detected moving target, and finally judging the motion of the moving human body by analyzing information data. However, this method of collecting motion information is problematic: because the moving image is shot in the dynamic scene, the moving image is influenced by factors such as on-site illumination change, moving object shielding and the like, and the moving image is easily interfered by background noise in the process of extracting and dividing the moving human body, so that the inaccurate moving contour dividing result of the human body is caused. In addition, to obtain the motion data parameters of the human body, only video streams of different angles during the motion of the human body can be combined with a related equipment calibration test and a specific calculation model to indirectly obtain the human body motion parameters (including joint and skeleton space postures, acceleration, speed, displacement, joint rotation angular speed and the like during the motion) and the dynamics parameters (including force, moment, acting and the like), the defects are obvious, firstly, the motion and dynamics information during the motion of the human body cannot be directly obtained, and the finally obtained precision of the human body motion data information is excessively dependent on the position and angle placed by the capturing equipment and the model precision required by conversion; secondly, the capturing device based on high-speed shooting has the defects of complex acquisition process and design, fixed visual angle, requirement for multi-view signal support and the like, so that the problems of high cost, poor portability, low signal dimension, low precision and the like of the acquisition device are caused. Accordingly, inertial sensor-based motion analysis and recognition is presented herein.
In the prior art, common recognition methods include a neural network, a support vector machine, a hidden Markov model and the like, and although the base recognition can be performed, the recognition is not visual enough, and the equipment collects a large amount of data in real time, so that the system needs to perform operation of mass data, the redundancy of the data in the system is easy to cause, and the response speed and the recognition accuracy of the system are reduced.
Disclosure of Invention
Aiming at the defects of the prior art, the invention solves the technical problems by adopting the following technical scheme: the AI-based motion recognition system comprises a sensor module, a receiving module, an acquisition module, a singlechip, an extraction module, a processing module and a recognition module, wherein the sensor module is used for detecting motion data of a human body through a sensor and sending the detected motion data to the receiving module through Bluetooth;
the receiving module is used for receiving the motion data sent by the sensor module, digitally encoding the received motion data and then transmitting the encoded motion data to the acquisition module through a wire;
the acquisition module is used for converting the measured signals into signals which can be identified by the singlechip and inputting the signals into the singlechip;
the singlechip is used for executing an algorithm program for testing;
the extraction module is used for receiving data information sent from the singlechip, extracting and classifying the received motion data, storing the classified data in the system, and calling the data from the system according to the requirement;
the processing module is used for processing the extracted data sent out by the extracting module;
the identification module is used for receiving the measured value sent from the processing module and retrieving the parameter from the database.
As a preferred embodiment, the sensor module collects the motion gesture signal of the human body by detecting force, speed, angle and height of the built-in sensor, wherein the force detected by the sensor is pressure, the detected speed is the angular speed and linear speed of the motion of the human body, the detected angle is the inclination angle of the human body relative to the ground, and the detected height is the vertical height of the human body relative to the ground.
As a preferred implementation mode, the acquisition module utilizes a singlechip to perform multichannel real-time acquisition, the acquisition frequency is 100HZ, and the acquisition actions are upward, downward, leftward, rightward and rotation.
As a preferred embodiment, the initial state of the human body posture is that the standing hands are unfolded and flattened, and six a-type sensors are worn on the human body, and the initial state, X, Y, Z, has three axes with values as origin points, namely (0, 0).
In a preferred embodiment, the extracting module extracts the features by fourier transform, and uses discrete time signalsIs defined as:
as a preferred embodiment, the values detected by the sensor module may be used to perform motion analysis on the human body by using an attitude angle, and the values are represented by a rotation angle of the sensor's own coordinate system relative to the northeast coordinate system:
heading angle: the angle of rotation of the sensor chip around the Z axis is defined, and the angle is defined as the angle between the projection of the longitudinal axis direction of the chip on the horizontal plane and the meridian line of the earth (namely the longitude line), namely the angle between the projection of the Y (b) axis of the carrier coordinate system on the horizontal plane of the northeast coordinate system and the Y (n) in the northeast coordinate system, the clockwise rotation direction is positive from the geographic north pole direction, and the range is 0-360 degrees;
pitch angle: the angle of rotation of the sensor chip around the X axis is defined, and the angle is defined as the angle between the axis Y (b) of the longitudinal axis of the chip and the geographic horizontal plane in space, namely the angle between the longitudinal axis Y (b) of the carrier and the projection of the carrier on the horizontal plane of the northeast coordinate system, the head-up of the chip is positive, the head-down is negative, and the range is [ -90 DEG ];
roll angle: the angle defined as the rotation angle of the sensor chip around the Y axis is the included angle between the horizontal axis X (b) of the chip and the horizontal plane in space, namely the included angle between the horizontal axis X (b) of the chip and the projection of the chip on the horizontal plane of the northeast coordinate system, the right side of the chip is lifted to be positive, the left side is lifted to be negative, and the definition range is [ -180 ° ].
As a preferred embodiment, the data acquisition module needs to preprocess the data and then extract and identify the data,
setting a threshold value, subtracting the acquired data from the original data, if the result is smaller than the threshold value, indicating that the acquisition is normal, otherwise, indicating that the acquisition is abnormal, and removing the data, wherein the removing method comprises the following steps of:
one-time median value: data sequenceTo->A new data sequence generated on the basis +.>,The construction method of (2) is as follows, from->Extract data->Taking the number of the middle digits and recording the number as +.>Then take out +.>Add->Constitute new data->Taking the number of the middle digits and recording as +.>Repeating the above steps until +_>To which the last element of (a) is added;
secondary median value: in the same way willThe median constituent sequence is extracted from the adjacent three data>;/>
Judging abnormal data: from the following componentsThe sequence is constructed according to the following formula>,
Will->-/>Comparing with the threshold value set previously, if the data is larger than the threshold value, the data is abnormal data.
As a preferred embodiment, the data acquisition module performs subsequent feature recognition and selection after preprocessing the data, where feature recognition and selection are description and characterization of recognition objects, extracts and selects feature vectors from original information, and extracts time domain features by using a probability statistical method, where the extracting includes:
as a preferred embodiment, the acquisition module performs preprocessing on the data received from the receiving unit, where the data includes acceleration of X, Y, Z three axes, converts the initial three-axis acceleration data into vector magnitude data, and obtains vector magnitude data by using euclidean norms, and the operations are as follows:the method is used for reducing interference of rotation component data in the triaxial acceleration data on motion recognition;
the acquisition module classifies and trains the acquired human body data by utilizing a convolutional neural network, and the training steps are as follows:
a. input: x, Y, Z triaxial acceleration data are input into the system after vector amplitude conversion;
b. convolution: a two-dimensional convolution operation is employed, specifically as follows,a represents the upper layer matrix, K () represents the convolution kernel, m x n represents the convolution kernel size, +.>Representing the magnitude of the convolved input,b represents the characteristic output after convolution;
c. and (3) removing correlation characteristics: for complex human body motion, the wearable sensor has strong characteristic correlation, the convolution of the neural network is specifically to calculate the weighted sum of the upper layer, the sequential operation of a plurality of convolution kernels can obtain a new two-dimensional matrix, the data volume of initial data is smaller, the acquisition items which are autonomously learned by the system are limited, the data can generate deviation, the phenomenon of fitting is generated, the dependency relationship among the numerical values can be transmitted along with the network, the next convolution layer is influenced, the covariance matrix is utilized to eliminate the correlation, the convolution layer is reconstructed, the inverse square root reciprocal calculation of the Newton-Lapherson algorithm is utilized, the inverse correction is adopted to reduce the correlation among the data, and the method is specifically calculated as follows:
the approximate inverse square of the covariance matrix D is calculated,where X is the data matrix, N is the number of samples, cov is a matrix representation of the covariance between the random tensors, applying the covariance matrix in the convolution process, multiplying the processed data by the convolution kernel W, and constructing the whole convolution layer by matrix multiplication to eliminate correlation.
The beneficial effects of the invention are as follows:
before the use, the tester wears all kinds of sensors at the A point of the body, and the sensors are fixed at relevant parts of the human body to form nodes for monitoring, when action data acquisition is carried out, in order to continuously record the action information of the tester, the capturing equipment needs to acquire at the frequency of 100HZ, and the frequency can not take excessive data in a short time on the premise of ensuring the adoption precision, so that the accuracy of motion analysis is improved, the human body action is limited according to the actual condition of the human body motion, not all limb space changes are human body actions, the actions have periodicity or obvious changes, on the basis, the AI can automatically screen limb actions without periodicity or obvious changes, the data quantity of the acquired data is further reduced, and the problem of slow judgment caused by overlarge acquired data of the system can be reasonably avoided.
The method is characterized in that a distributed database is adopted to upload data, the acquired data form a set by a group of documents, fields can be added or deleted for any one or a group of documents, and the distributed expansion is realized by utilizing an automatic slicing mechanism.
The data are used as a bottom data stream, stored in the system and are the judgment basis of AI learning, each data are sent to the system in real time in the subsequent human body wearing and collecting process, AI realizes complex inquiry from the bottom data stream, a large number of data streams are counted, screened and segmented, and then the data are reversely pushed to a set through the blocks, and the human body posture is judged by utilizing the characteristic of the set data.
Drawings
FIG. 1 is a system block diagram of the present invention;
FIG. 2 is a data block diagram of a sensor module and a receiving module of the present invention;
FIG. 3 is a flow chart of the data stream conversion of the present invention;
FIG. 4 is a schematic illustration of the wearing of the present invention by a wearer;
FIG. 5 is a graph of real-time monitoring data of the present invention;
FIG. 6 is a graph of the results of the verification of the present invention.
Detailed Description
The invention will be described in further detail with reference to the drawings and the detailed description. The embodiments of the invention have been presented for purposes of illustration and description, and are not intended to be exhaustive or limited to the invention in the form disclosed. Many modifications and variations will be apparent to those of ordinary skill in the art. The embodiments were chosen and described in order to best explain the principles of the invention and the practical application, and to enable others of ordinary skill in the art to understand the invention for various embodiments with various modifications as are suited to the particular use contemplated.
Examples:
referring to fig. 1 to 6, the present invention provides a technical solution: the AI-based motion recognition system comprises a sensor module, a receiving module, an acquisition module, an extraction module, a processing module and a recognition module, wherein the sensor module is used for detecting motion data of a human body through a sensor and sending the detected motion data to the receiving module through Bluetooth; the receiving module is used for receiving the motion data sent by the sensor module, digitally encoding the received motion data and then transmitting the encoded motion data to the acquisition module through a wire; the acquisition module converts the measured signals into signals which can be identified by the singlechip and inputs the signals to the singlechip; the extraction module receives data information sent by the singlechip, extracts and classifies the received motion data, stores the classified data in the system, and retrieves the data from the system according to the requirement; the processing module is used for obtaining a measured value corresponding to the measured parameter after the singlechip executes an algorithm program for testing; the identification module is used for receiving the measured value sent from the processing module and retrieving the parameter from the database.
The sensor module collects human body movement posture signals by detecting force, speed, angle and height, the detected force is pressure, the detected speed is angular speed and linear speed of human body movement, the detected angle is inclination angle of the human body relative to the ground, and the detected height is vertical height of the human body relative to the ground. The acquisition module utilizes the singlechip to carry out multichannel real-time acquisition, the acquisition frequency is 100HZ, and the actions of acquisition are upward, downward, leftward, rightward and turning. The initial state of the human body posture is that the standing hands are unfolded and unfolded, six A-type sensors are worn on the human body, and the numerical values of the X, Y, Z three axes in the initial state are the origin points, namely (0, 0).
In the extraction module, the Fourier transform is adopted to extract the characteristics, and the discrete time signals are obtainedIs defined as:
the numerical value detected by the sensor module can be used for carrying out motion analysis on a human body by using an attitude angle, and the rotation angle of the sensor self coordinate system relative to the northeast day coordinate system is represented by:
heading angle: the angle of rotation of the sensor chip around the Z axis is defined, and the angle is defined as the angle between the projection of the longitudinal axis direction of the chip on the horizontal plane and the meridian line of the earth (namely the longitude line), namely the angle between the projection of the Y (b) axis of the carrier coordinate system on the horizontal plane of the northeast coordinate system and the Y (n) in the northeast coordinate system, the clockwise rotation direction is positive from the geographic north pole direction, and the range is 0-360 degrees;
pitch angle: the angle of rotation of the sensor chip around the X axis is defined, and the angle is defined as the angle between the axis Y (b) of the longitudinal axis of the chip and the geographic horizontal plane in space, namely the angle between the longitudinal axis Y (b) of the carrier and the projection of the carrier on the horizontal plane of the northeast coordinate system, the head-up of the chip is positive, the head-down is negative, and the range is [ -90 DEG ];
roll angle: the angle defined as the rotation angle of the sensor chip around the Y axis is the included angle between the horizontal axis X (b) of the chip and the horizontal plane in space, namely the included angle between the horizontal axis X (b) of the chip and the projection of the chip on the horizontal plane of the northeast coordinate system, the right side of the chip is lifted to be positive, the left side is lifted to be negative, and the definition range is [ -180 ° ].
The data acquisition module needs to preprocess the data and then extract and identify the data,
setting a threshold value, subtracting the acquired data from the original data, if the result is smaller than the threshold value, indicating that the acquisition is normal, otherwise, indicating that the acquisition is abnormal, and removing the data, wherein the removing method comprises the following steps of:
one-time median value: data sequenceTo->A new data sequence generated on the basis +.>,The construction method of (2) is as follows, from->Extract data->Taking the number of the middle digits and recording the number as +.>Then take out +.>Add->Constitute new data->Taking the number of the middle digits and recording as +.>Repeating the above steps until +_>To which the last element of (a) is added;
secondary median value: by the same methodThe method willThe median constituent sequence is extracted from the adjacent three data>;
Judging abnormal data: from the following componentsThe sequence is constructed according to the following formula>,/>
Will->-/>Comparing with the threshold value set previously, if the data is larger than the threshold value, the data is abnormal data.
The data acquisition module carries out preprocessing on the data and then carries out subsequent feature recognition and selection, wherein the feature recognition and selection are description and characterization of recognition objects, feature vectors are extracted and selected from original information, a probability statistical method is adopted to extract time domain features, and the method comprises the following steps:
motion characteristics of motion recognition are basically embodied on the trunk and the legs, redundant characteristic information is removed by extracting time domain characteristic information, system load is reduced, and response speed and recognition accuracy are improved.
Before using, the tester wears various sensors at the A point of the body, the sensors are fixed at the relevant parts of the human body to form nodes for monitoring, when collecting action data, in order to continuously record the action information of the tester, the capturing equipment needs to collect at the frequency of 100HZ, the frequency can ensure that excessive data can not be adopted in a short time under the premise of adopting precision, a distributed database is adopted for uploading the data, the collected data form a set by a group of documents, the fields can be added or deleted to any one or a group of documents, the automatic slicing mechanism is utilized for realizing distributed expansion, the principle is that the set is divided into small blocks, the blocks can be dispersed on different slices, each slice is responsible for a part of the set, the data is automatically stored on the database nodes, the initial sampling data is sent into the database, each action forms a specified action by the set, and the action forms are defined by collecting the heading angle, the pitch angle, the rolling angle and the height.
The data are used as a bottom data stream, stored in the system and are the judgment basis of AI learning, each data are sent to the system in real time in the subsequent human body wearing and collecting process, AI realizes complex inquiry from the bottom data stream, a large number of data streams are counted, screened and segmented, and then the data are reversely pushed to a set through the blocks, and the human body posture is judged by utilizing the characteristic of the set data.
In order to improve the accuracy of motion analysis, the human body motion is limited according to the actual condition of human body motion, not all limb space changes are human body motions, the motions have periodicity or obvious changes, on the basis, the AI can automatically screen limb motions without periodicity or obvious changes, the data volume of acquired data is further reduced, and the problem of slow judgment caused by overlarge acquired data of a system can be reasonably avoided.
It will be apparent that the described embodiments are only some, but not all, embodiments of the invention. All other embodiments, which can be made by those skilled in the art and which are included in the embodiments of the present invention without the inventive step, are intended to be within the scope of the present invention. Structures, devices and methods of operation not specifically described and illustrated herein are implemented by conventional means in the art unless specifically indicated and limited by the context.
Claims (9)
1. The utility model provides a motion recognition system based on AI, includes sensor module, receiving module, collection module, singlechip, draws module, processing module and recognition module, its characterized in that: the sensor module is used for detecting motion data of a human body through a sensor and sending the detected motion data to the receiving module through Bluetooth;
the receiving module is used for receiving the motion data sent by the sensor module, digitally encoding the received motion data and then transmitting the encoded motion data to the acquisition module through a wire;
the acquisition module is used for converting the measured signals into signals which can be identified by the singlechip and inputting the signals into the singlechip;
the singlechip is used for executing an algorithm program for testing;
the extraction module is used for receiving data information sent from the singlechip, extracting and classifying the received motion data, storing the classified data in the system, and calling the data from the system according to the requirement;
the processing module is used for processing the extracted data sent out by the extracting module;
the identification module is used for receiving the measured value sent from the processing module and retrieving the parameter from the database.
2. The AI-based motion recognition system of claim 1, wherein: the sensor module collects human body movement posture signals through built-in sensors, the force, the speed, the angle and the height are measured, the force measured by the sensors is pressure, the measured speed is the angular speed and the linear speed of human body movement, the measured angle is the inclination angle of the human body relative to the ground, and the measured height is the vertical height of the human body relative to the ground.
3. The AI-based motion recognition system of claim 1, wherein: the acquisition module utilizes a singlechip to carry out multichannel real-time acquisition, the acquisition frequency is 100HZ, and the acquisition actions are upward, downward, leftward, rightward and turning.
4. The AI-based motion recognition system of claim 1, wherein: the initial state of the human body posture is that the standing hands are unfolded and flattened, six A-type sensors are worn on the human body, and the numerical values of the X, Y, Z three axes in the initial state are the origin points, namely (0, 0).
6. the AI-based motion recognition system of claim 1, wherein: the numerical value detected by the sensor module can be used for carrying out motion analysis on a human body by using an attitude angle, and the rotation angle of the sensor self coordinate system relative to the northeast coordinate system is represented by:
heading angle: the angle of rotation of the sensor chip around the Z axis is defined, and the angle is defined as the angle between the projection of the longitudinal axis direction of the chip on the horizontal plane and the meridian line of the earth, namely the angle between the projection of the Y (b) axis of the carrier coordinate system on the horizontal plane of the northeast coordinate system and the Y (n) axis of the northeast coordinate system, the clockwise rotation direction is positive from the geographic north pole direction, and the range is 0-360 degrees;
pitch angle: the angle of rotation of the sensor chip around the X axis is defined, and the angle is defined as the angle between the axis Y (b) of the longitudinal axis of the chip and the geographic horizontal plane in space, namely the angle between the longitudinal axis Y (b) of the carrier and the projection of the carrier on the horizontal plane of the northeast coordinate system, the head-up of the chip is positive, the head-down is negative, and the range is [ -90 DEG ];
roll angle: the angle defined as the rotation angle of the sensor chip around the Y axis is the included angle between the horizontal axis X (b) of the chip and the horizontal plane in space, namely the included angle between the horizontal axis X (b) of the chip and the projection of the chip on the horizontal plane of the northeast coordinate system, the right side of the chip is lifted to be positive, the left side is lifted to be negative, and the definition range is [ -180 ° ].
7. The AI-based motion recognition system of claim 1, wherein: the data acquisition module needs to preprocess the data and then extract and identify the data,
setting a threshold value, subtracting the acquired data from the original data, if the result is smaller than the threshold value, indicating that the acquisition is normal, otherwise, indicating that the acquisition is abnormal, and removing the data, wherein the removing method comprises the following steps of:
one-time median value: data sequenceTo->A new data sequence generated on the basis +.>,/>The construction method of (2) is as follows, from->Extract data->Taking the number of the middle digits and recording the number as +.>Then take out +.>Add->Constitute new data->Taking the number of the middle digits and recording as +.>Repeating the above steps until +_>To which the last element of (a) is added;
secondary median value: in the same way willThe median constituent sequence is extracted from the adjacent three data>;
Judging abnormal data: from the following componentsThe sequence is constructed according to the following formula>,
8. The AI-based motion recognition system of claim 1, wherein: the data acquisition module performs preprocessing on the data, and then performs subsequent feature recognition and selection, wherein the feature recognition and selection are description and characterization of recognition objects, feature vectors are extracted and selected from original information, a probability statistical method is adopted to extract time domain features, and the method comprises the following steps:
9. the AI-based motion recognition system of claim 1, wherein: the acquisition module is used for preprocessing the data received from the receiving module, wherein the data comprise acceleration of X, Y, Z three axes, the initial three-axis acceleration data are converted into vector magnitude data, and the vector magnitude data are obtained by utilizing Euclidean norms, and the operations are as follows:;
the acquisition module classifies and trains the acquired human body data by utilizing a convolutional neural network, and the training steps are as follows:
a. input: x, Y, Z triaxial acceleration data are input into the system after vector amplitude conversion;
b. convolution: a two-dimensional convolution operation is employed, specifically as follows,a represents the upper layer matrix, K () represents the convolution kernel, m x n represents the convolution kernel size, +.>Representing the magnitude of the convolved input,b represents the characteristic output after convolution;
c. and (3) removing correlation characteristics: for complex human body motion, the wearable sensor has strong characteristic correlation, the convolution of the neural network is specifically to calculate the weighted sum of the upper layer, the sequential operation of a plurality of convolution kernels can obtain a new two-dimensional matrix, the data volume of initial data is smaller, the acquisition items which are autonomously learned by the system are limited, the data can generate deviation, the phenomenon of fitting is generated, the dependency relationship among the numerical values can be transmitted along with the network, the next convolution layer is influenced, the covariance matrix is utilized to eliminate the correlation, the convolution layer is reconstructed, the inverse square root reciprocal calculation of the Newton-Lapherson algorithm is utilized, the inverse correction is adopted to reduce the correlation among the data, and the method is specifically calculated as follows:
the approximate inverse square of the covariance matrix D is calculated,where X is the data matrix, N is the number of samples, cov is a matrix representation of the covariance between the random tensors, applying the covariance matrix in the convolution process, multiplying the processed data by the convolution kernel W, and constructing the whole convolution layer by matrix multiplication to eliminate correlation. />
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310491790.7A CN116226727A (en) | 2023-05-05 | 2023-05-05 | Motion recognition system based on AI |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310491790.7A CN116226727A (en) | 2023-05-05 | 2023-05-05 | Motion recognition system based on AI |
Publications (1)
Publication Number | Publication Date |
---|---|
CN116226727A true CN116226727A (en) | 2023-06-06 |
Family
ID=86569765
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202310491790.7A Pending CN116226727A (en) | 2023-05-05 | 2023-05-05 | Motion recognition system based on AI |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN116226727A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117574133A (en) * | 2024-01-11 | 2024-02-20 | 湖南工商大学 | Unsafe production behavior identification method and related equipment |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104237687A (en) * | 2014-09-12 | 2014-12-24 | 华中科技大学 | On-line monitoring method for ageing service life of energy supply laser of active electronic transformer |
CN113642432A (en) * | 2021-07-30 | 2021-11-12 | 南京师范大学 | Method for identifying human body posture by convolutional neural network based on covariance matrix transformation |
CN116027905A (en) * | 2023-01-18 | 2023-04-28 | 大连理工大学 | Double kayak upper limb motion capturing method based on inertial sensor |
-
2023
- 2023-05-05 CN CN202310491790.7A patent/CN116226727A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104237687A (en) * | 2014-09-12 | 2014-12-24 | 华中科技大学 | On-line monitoring method for ageing service life of energy supply laser of active electronic transformer |
CN113642432A (en) * | 2021-07-30 | 2021-11-12 | 南京师范大学 | Method for identifying human body posture by convolutional neural network based on covariance matrix transformation |
CN116027905A (en) * | 2023-01-18 | 2023-04-28 | 大连理工大学 | Double kayak upper limb motion capturing method based on inertial sensor |
Non-Patent Citations (3)
Title |
---|
侯祖贵: "基于惯性传感器的人体动作分析与识别", 中国优秀硕士学位论文全文数据库信息科技辑, vol. 2014, no. 04, pages 6 - 50 * |
权威铭 等: "带有协方差矩阵的卷积神经网络在人体运动识别中的应用", 计算机工程与科学, vol. 44, no. 11, pages 2029 - 2031 * |
杨丽娟 等: "快速傅里叶变换FFT及其应用", 光电工程, vol. 31, pages 2 * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117574133A (en) * | 2024-01-11 | 2024-02-20 | 湖南工商大学 | Unsafe production behavior identification method and related equipment |
CN117574133B (en) * | 2024-01-11 | 2024-04-02 | 湖南工商大学 | Unsafe production behavior identification method and related equipment |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107220617A (en) | Human body attitude identifying system and method | |
CN108305283B (en) | Human behavior recognition method and device based on depth camera and basic gesture | |
CN108629946B (en) | Human body falling detection method based on RGBD sensor | |
CN110334573B (en) | Human motion state discrimination method based on dense connection convolutional neural network | |
Li et al. | Intelligent sports training system based on artificial intelligence and big data | |
CN105160310A (en) | 3D (three-dimensional) convolutional neural network based human body behavior recognition method | |
US9183431B2 (en) | Apparatus and method for providing activity recognition based application service | |
CN108245172B (en) | Human body posture recognition method free of position constraint | |
JP2014522035A (en) | Object posture search apparatus and method | |
CN103324938A (en) | Method for training attitude classifier and object classifier and method and device for detecting objects | |
CN104408718B (en) | A kind of gait data processing method based on Binocular vision photogrammetry | |
CN110659677A (en) | Human body falling detection method based on movable sensor combination equipment | |
CN111274998A (en) | Parkinson's disease finger knocking action identification method and system, storage medium and terminal | |
CN112200162A (en) | Non-contact heart rate measuring method, system and device based on end-to-end network | |
CN107122711A (en) | A kind of night vision video gait recognition method based on angle radial transformation and barycenter | |
CN116226727A (en) | Motion recognition system based on AI | |
CN111914643A (en) | Human body action recognition method based on skeleton key point detection | |
CN113378649A (en) | Identity, position and action recognition method, system, electronic equipment and storage medium | |
CN115859078A (en) | Millimeter wave radar fall detection method based on improved Transformer | |
Chen et al. | Detection of falls with smartphone using machine learning technique | |
CN108182410A (en) | A kind of joint objective zone location and the tumble recognizer of depth characteristic study | |
CN109190762A (en) | Upper limb gesture recognition algorithms based on genetic algorithm encoding | |
CN113350771A (en) | Athlete dynamic posture recognition method, device, system and storage medium | |
CN113033501A (en) | Human body classification method and device based on joint quaternion | |
CN110801227A (en) | Method and system for testing three-dimensional color block obstacle based on wearable equipment |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20230606 |
|
RJ01 | Rejection of invention patent application after publication |