CN109977881B

CN109977881B - Character action feature extraction and identification optimization method based on radio frequency technology

Info

Publication number: CN109977881B
Application number: CN201910245310.2A
Authority: CN
Inventors: 袁呈呈; 陈志�; 岳文静
Original assignee: Nanjing University of Posts and Telecommunications
Current assignee: Nanjing University of Posts and Telecommunications
Priority date: 2019-03-28
Filing date: 2019-03-28
Publication date: 2022-07-22
Anticipated expiration: 2039-03-28
Also published as: CN109977881A

Abstract

The invention discloses a figure action characteristic extraction and identification optimization method based on a radio frequency technology. The reader can send out signals to the space, and the tags can send out response signals after receiving the signals. A plurality of readers are arranged in a space, and tags are attached to the human body. When a user acts, signals received by the readers are changed, when the action is finished, data obtained by the readers are integrated, the change of the signals in the movement process is divided into a plurality of sections, and the type of the action of the user is identified by means of a deep learning technology. The method solves the problem of high computational power requirement of the video motion recognition technology.

Description

Figure action characteristic extraction and identification optimization method based on radio frequency technology

Technical Field

The invention relates to a figure action feature extraction and recognition optimization method, and belongs to the technical field of cross of Internet of things and machine learning. The action characteristics are obtained mainly by identifying a radio frequency technology, and then the action is identified by using a deep learning technology.

Background

As a fundamental problem in the field of computer vision, human motion recognition aims to enable a machine to semantically understand and analyze human motion videos through algorithms. Potential applications based on human body action recognition include a plurality of fields such as intelligent monitoring, video content analysis and behavior perception, and intelligent home. Due to such wide application scenes and potential application values, a large amount of scientific research work and research subjects are developed around human body action recognition in the field of computer vision. The potential application based on human body action recognition comprises multiple fields of intelligent monitoring, video content analysis, human body interaction, intelligent home furnishing and the like. Specifically, the research on human motion recognition mainly includes, in application classification: (1) intelligent monitoring; (2) analyzing the video content; (3) behavioral awareness and intelligent control.

In the action recognition system, the traditional behavior perception is completed through input devices such as a keyboard and a mouse, and the traditional behavior perception is a process that human beings adapt to machines. If the machine can be made to understand human languages, including natural languages and body languages of human beings, the efficiency and experience of behavior perception are greatly improved. The understanding of human body language by computer requires classification and recognition of human actions. In addition, if the human activities are recognized and understood through a camera or other machines, and the virtual world processes the information in feedback and virtual world, the machines can help people to complete tasks such as gesture recognition, gesture recognition and the like, correct action error areas of certain activities or skills, and also can entertain life and experience games of behavior perception.

The radio frequency identification technology is gradually permeating into modern life, and has been widely applied in industries such as retail consumer goods, medicine management, transportation, national defense and military industry and the like due to high efficiency and convenience brought by the radio frequency identification technology, and particularly, the unique identification characteristic of the radio frequency identification technology to objects stimulates the research interest of people on the internet of things.

Most of current action discernment still are action recognition technology based on vision, thereby obtain information with the camera admission video promptly, and the video has been as the main carrier of information, and the quantity of video data is explosive growth, and the content that all the time has a large amount of new produces. In the face of video data which is massively emerging, the main problem to be solved by action recognition is that a computer processes and analyzes original images or image sequence data acquired by a camera, learns and understands actions and behaviors of people in the images, generally, on the basis of motion detection and feature extraction, a human motion mode is obtained through analysis, and a mapping relation between video content and action type description is established, so that the computer can 'see' the video or 'understand' the video, namely, the method mainly comprises 3 steps: (1) detecting motion information from the image frame and extracting bottom layer characteristics; (2) modeling a behavior pattern or action; (3) and establishing a corresponding relation between the bottom visual characteristics and high-level semantic information such as action behavior lists.

Although the vision-based motion recognition technology is concerned by many colleges and enterprises at home and abroad, the fact that the storage amount of video information is large is a non-competitive fact, and the processing speed of the video cannot be effectively improved all the time due to the limitations of a processor and an algorithm.

Disclosure of Invention

The invention aims to: in order to overcome the defects in the prior art, the invention provides a person motion characteristic extraction and identification optimization method based on a radio frequency technology, and the method effectively solves the problem of high computational power requirement of a video motion analysis technology.

A character action feature extraction and identification method based on a radio frequency identification technology is disclosed, and a radio frequency identification system mainly comprises a reader and a tag. The reader can send out signals to the space, and the tags can send out response signals after receiving the signals. A plurality of readers are arranged in a space, and tags are attached to the human body. When a user acts, signals received by the readers are changed, when the action is finished, data obtained by the readers are integrated, the change of the signals in the movement process is divided into a plurality of sections, and the type of the action of the user is identified by means of a deep learning technology. The method is used for solving the problem of high computational power requirement of the video motion recognition technology, and specifically comprises the following steps:

step 1), attaching a label to a human body, enabling a user to act, and receiving and storing the change of the phase difference between a sent signal and a received signal by a plurality of readers in a space;

and 2) when the user action is finished, the system splices the phase difference data to obtain a phase difference change process.

And 3) equally dividing the phase difference change process into a plurality of sections.

And 4) summarizing the data of all readers.

Step 5), the processed data is delivered to a deep learning network, and the action type is judged;

step 6), controlling other software to make corresponding response according to the judged action type;

wherein the specific steps of the step 1) are as follows:

and 11) arranging m readers in the space.

Step 12) attaching the label to the human body.

And step 13), the system receives and stores the data of each reader.

Wherein the specific steps of the step 2) are as follows:

step 21), setting n items of original phase difference data received by a reader, and recording the n items as input1[ n ];

step 22), recording the spliced data as input2[ n ], wherein the splicing process comprises the following steps:

obtaining a processed data input2[ n ], the data identifying the variation information of the phase difference;

wherein the specific steps of the step 3) are as follows:

step 31), if the processed data has t items, which are recorded as input3 t, and j is more than or equal to 1 and less than or equal to t, the following processing is carried out

input3[t]T points are taken in the motion process, and the value of each point corresponds to the size of the phase difference change of the label in the motion process;

wherein the specific steps of the step 4) are as follows:

and step 41) processing the data obtained by the m readers according to the methods from the step 2) to the step 3). A total of mt data are obtained. The data are collectively recorded as data [ mt ] as a preprocessing result;

wherein, the specific steps of the step 6) are as follows:

step 61), firstly, d action instructions need to be recorded, and data preprocessed by the action instructions are obtained through the step 2)3) 4);

step 62), constructing an analog data generator, and generating scrambled data offset [ mt ] for a piece of real data [ mt ], wherein each term is a random number between (-r, r). Let sim [ mt ] + offset [ mt ] be the piece of analog data obtained. Generating a large amount of simulation data according to the method;

and step 63) building a deep learning network, wherein the network comprises an input layer 1 layer, mt nodes, a hidden layer 1 layer mt nodes, an output layer 1 layer and the number d of the nodes. The training of the deep learning network is completed by matching with a simulation data generator;

step 64), solidifying the trained deep learning network;

step 65), in the identification process, the deep learning network obtained from the step 61) to the step 64) is used for processing real data, and the action type of the real data is judged;

in the step 11), m is taken as 4 according to experience;

in the step 31), t is taken as 20 according to experience;

in the step 61), d is taken as 3 according to experience;

in the step 62), r is empirically 0.5;

has the advantages that: compared with the existing video motion recognition technology, the technical scheme adopted by the invention has the following technical effects:

(1) the invention obtains the action characteristics through the radio frequency identification, and the information quantity is small, thereby the identification is rapid.

(2) The invention obtains the action characteristics through the radio frequency identification, and has less information quantity, thereby having low calculation force requirement.

Drawings

FIG. 1 is a flow chart of a method for character motion feature extraction and recognition optimization;

FIG. 2 preprocesses the resulting data;

Detailed Description

The present invention is further illustrated in the accompanying drawings and described in the following detailed description, it is to be understood that such examples are included solely for the purposes of illustration and are not intended as a definition of the limits of the invention, since various equivalent modifications of the invention will become apparent to those skilled in the art after reading the present specification, and it is intended to cover all such modifications as fall within the scope of the invention as defined in the appended claims.

The invention discloses a character action feature extraction and identification method based on a radio frequency identification technology, which comprises the following steps as shown in figures 1 to 2:

step 1), 4 readers are arranged in space, and the tag is attached to a human body. As the user makes an action, the system receives and stores data for each reader.

And step 2) when the user action is finished, the system splices the phase difference data to obtain a phase difference change process. Setting n items of original phase difference data received by a reader, and recording as input1[ n ];

recording the spliced data as input2[ n ], wherein the splicing process comprises the following steps:

obtaining a processed data input2[ n ], which identifies the variation information of the phase difference;

and 3) equally dividing the phase difference change process into a plurality of sections. Recording the processed data as input3[20], and for j being more than or equal to 1 and less than or equal to 20, processing as follows:

input3[20]the method is characterized in that 20 points are taken in the process of movement, and the value of each point corresponds to the phase change of the label in the process of the movement;

and 4) summarizing data of all readers. And processing the data obtained by the 4 readers according to the methods from the step 2) to the step 3). A total of 80 data were obtained. The data are collectively recorded as data [80] as a result of preprocessing;

step 6), controlling other software to make corresponding response according to the judged action type; firstly, inputting 3 action instructions, and acquiring data preprocessed by the action instructions through the step 2)3) 4); an analog data generator is then constructed to generate scrambled data offsets [80] for a piece of real data [80], each term being a random number between (-0.5, 0.5). Let sim [80] + data [80] be the piece of analog data obtained. Generating a large amount of simulation data according to the method; and then constructing a deep learning network, wherein the network comprises an input layer 1 layer, 80 nodes, a hidden layer 1 layer, 80 nodes, an output layer 1 layer and 3 nodes. The training of the deep learning network is completed by matching with a simulation data generator; solidifying the trained deep learning network; in the identification process, the obtained deep learning network is used for processing real data and judging the action type of the real data;

the invention is further described below with reference to the accompanying drawings:

in specific implementation, fig. 1 is a flow chart of a human action feature extraction and identification method based on a radio frequency identification technology.

First 4 readers are placed at different locations in space and then tags are attached to the person's hand positions. The reader sends a signal to the space, and the tag sends a response signal after receiving the signal. The reader may compare the phase difference between the own transmitted signal and the received reply signal.

In the process of movement of a user, the tag moves along with the movement of the user, the distance between the tag and the reader is continuously changed, the phase difference is also changed, and the system stores the phase difference obtained each time.

When the user action is finished, the phase difference stored by one reader is recorded as input1[ n ].

Is processed as follows

The first round of processed data input2[ n ] is obtained, which identifies the variation information of the phase difference.

And then the following treatment is carried out:

data input3 was obtained after the second round of processing [20 ]. And processing and summarizing the data acquired by the 4 readers one by one according to the method to obtain a preprocessing result data [80 ].

The resulting data is preprocessed once as shown in fig. 2.

And then, judging the data [80] by utilizing a pre-trained and cured deep network model to obtain the action type.

The above description is only of the preferred embodiments of the present invention, and it should be noted that: it will be apparent to those skilled in the art that various modifications and adaptations can be made without departing from the principles of the invention and these are intended to be within the scope of the invention.

Claims

1. A character action feature extraction and identification optimization method based on a radio frequency technology is characterized by comprising the following steps:

step 1), attaching a tag to a human body, arranging m readers in a space, enabling a user to act, enabling the readers to read tag information, enabling a system to receive and store data of each reader to obtain tag phase difference data, and enabling the system to send and receive a signal phase difference;

step 2), when the user action is finished, the system splices the phase difference data to obtain a phase difference change process, and the steps are as follows:

step 21), the original phase difference data received by the reader has n items in total, and the n items are recorded as input1[ n ];

step 22), the spliced data is input2[ n ], and the splicing process is as follows:

obtaining a processed data input2[ n ];

step 3), the data processed in the step 2 has t items, which are marked as input3[ t ], and j is more than or equal to 1 and less than or equal to t, the following processing is carried out:

input3[t]the method comprises the following steps of (1) taking t points in the motion process, wherein the value of each point corresponds to the change magnitude of a phase difference of a tag in the motion process;

step 4), summarizing data of all readers;

and 5) delivering the data processed in the step 4) to a deep learning network, and judging the action type.

2. The method for extracting and identifying character motion features based on the wireless radio frequency technology as claimed in claim 1, wherein the method comprises the following steps: the step 4) of summarizing the data of all readers is as follows: processing the data obtained by the m readers according to the methods from the step 2) to the step 3), obtaining mt data in total, and summarizing the data as data [ mt ], wherein the data is used as a preprocessing result.

3. The method for extracting and identifying character motion features based on wireless radio frequency technology as claimed in claim 2, wherein: the specific steps of the step 5) are as follows:

step 51), d action instructions are required to be input, and data preprocessed by the action instructions are obtained through the step 2)3) 4);

step 52), constructing an analog data generator, generating scrambled data offset [ mt ] for a piece of real data [ mt ], wherein each term is a random number between (-r, r), r represents a random number determination interval value, sim [ mt ] ═ data [ mt ] + offset [ mt ] is used as an obtained piece of analog data sim [ mt ], and generating analog data according to the method;

step 53), building a deep learning network, wherein the network comprises an input layer 1 layer, mt nodes, a hidden layer 1 layer, mt nodes, an output layer 1 layer and node number d, and matching with a simulation data generator to finish the training of the deep learning network;

step 54), solidifying the trained deep learning network;

step 55), in the identification process, the deep learning network obtained in the step 51) to the step 54) is used for processing the real data and judging the action type of the real data.

4. The method for extracting and identifying character motion features based on wireless radio frequency technology as claimed in claim 3, wherein: the number m of the readers is 4.

5. The method for extracting and identifying character motion features based on wireless radio frequency technology as claimed in claim 4, wherein: the number of points t taken during the exercise is taken as 20.

6. The method for extracting and identifying character motion features based on the wireless radio frequency technology as claimed in claim 5, wherein the method comprises the following steps: the random number determination interval value r takes 0.5.