CN106648068A - Method for recognizing three-dimensional dynamic gesture by two hands - Google Patents

Abstract

The present invention proposes a two-hands three-dimensional dynamic gesture recognition method and system, using two mobile terminals, the two mobile terminal devices adopt the master-slave mode, one of them is used as the master mobile terminal, and the other is used as the slave mobile terminal; The mobile terminal tracks the gestures of the left hand and the right hand respectively; the master mobile terminal scans the external devices that can be connected to the outside world, and establishes a wireless data connection with the slave mobile terminal; the master mobile terminal collects gesture data; The gesture data is sent to the main mobile terminal in real time for data processing, synchronization and two-hand gesture recognition; the main mobile terminal sends the recognition result between the main mobile terminal and the gesture data obtained from the mobile terminal to the device to be controlled. The present invention directly uses the human hand as the input terminal to the smart device, and the interaction between the human and the machine does not need other media, which improves the efficiency of the interaction and avoids various misidentifications caused by the intermediate media.

Description

A three-dimensional dynamic gesture recognition method for both hands

technical field

The invention relates to the technical field of human-computer interaction, in particular to a dynamic gesture recognition method and system.

Background technique

With the advancement of technology in all aspects, people hope to get a better human-computer interaction experience, and the requirements for hard indicators such as real-time interaction and recognition rate are becoming more and more stringent. Information, sound wave information, eyeball information, brain wave information, muscle information, etc. to achieve communication with the machine. This has prompted many experts and scholars to constantly search for more convenient, faster, more accurate, and more natural human-computer interaction methods, and began to pay attention to the use of human body language to achieve machine control, among which gestures are the most varied and most capable. One of the ways to express people's wishes has become the focus of attention. Before the era of brain waves controlling the world, gestures are the research entry point that is most likely to change the traditional way of human-computer interaction. Because the hands are the most changeable and expressive item of body posture, they can express people's thoughts and wishes to the greatest extent possible. Directly using a simple gesture to control life without any medium is the ideal way of human-computer interaction that people are currently pursuing. This brand-new operation method enables human beings to truly control their own lives with their own hands in this era of interconnected intelligence where everything is connected. Gestures will also expand the human-computer interaction method from the traditional two-dimensional recognition technology to three-dimensional spatial recognition technology, realizing more possibilities for technological development. The two-hands three-dimensional dynamic gesture recognition technology based on the IOS mobile terminal to be done in the present invention uses the IOS intelligent mobile terminal which is more and more popular in people's life directly as a tool for gesture collection and recognition to realize friendly human-computer interaction with the PC terminal or other terminals Way.

Several mainstream solutions have been produced during the development of gesture recognition technology, such as:

1) The solution based on data gloves: the earliest start and the most direct solution to the problem. This scheme has many advantages, such as small amount of data, high speed, less affected by the environment, and the ability to directly acquire gesture data. However, users are required to wear complicated gloves, which greatly reduces the user experience, and it is technically difficult to coat the sensors attached to them with a relatively hard protective layer, and relatively fragile devices are also inconvenient to carry.

2) Processing solutions based on machine vision: the most popular gesture recognition technology solutions, that is, camera image processing, such as Microsoft's Kinect. The biggest advantage is that the user does not need to wear any equipment, and the human-computer interaction is good. However, this technology has relatively large restrictions on space and environment, so except for some large-scale somatosensory games, the promotion degree in life is not high.

3) The SEMG-based solution: In a multidisciplinary field, in addition to requiring the user to wear electrodes like the data glove solution, similar to the machine vision solution, this solution is also susceptible to interference from factors such as the external environment and users. The interference of the external environment includes the interference of the electromagnetic environment and the change of the contact resistance of the measuring electrode caused by the change of the environmental humidity. At the same time, the user's physical and mental state has different benchmark values in different situations, making it difficult to calibrate the system.

The existing human-computer interaction technology has the following disadvantages:

(1) Mechanical keyboard interaction is the most mature, but at the same time, the least efficient way of interaction. Due to various reasons such as cost, loss, and inefficiency, it is being gradually eliminated by mobile devices;

(2) Touch screen operation, as an emerging human-computer interaction method in recent years, has become an effective interaction method for a new generation of mobile devices due to its advantages of simple operation, user habit, low learning cost, and multi-touch. However, the touch technology inevitably requires the user to operate on the touch panel, which limits the application scenarios and scope of the touch technology.

(3) Speech recognition is currently a very mature interaction method with a high recognition rate, especially with the emergence of Siri in the iPhone, the recognition technology based on natural speech has brought the speech interaction method to a new level. However, the voice interaction method will be affected by regional language differences and personal voice differences. At the same time, natural voice recognition needs to be connected to the Internet at all times and has high energy consumption, which greatly limits the application of voice technology in wearable devices.

In view of the above problems, the inventor of this case proposed a fast three-dimensional dynamic gesture recognition method and system based on the subdivision eigenvalue method in Patent Document 1 (Chinese Patent Publication No. CN105929940A), aiming at the problems existing in the prior art, the design A small-looking hardware platform is built, using an accelerometer to collect the 3-axis acceleration of the user gesture, a gyroscope to collect the 3-axis angular velocity of the user gesture, a magnetometer to collect the 3-axis magnetic induction intensity of the user gesture, and a filtering algorithm to eliminate data errors Afterwards, the attitude calculation is carried out to obtain the real-time three attitude angles (heading angle, roll angle, and pitch angle), and then these information data are passed through the proposed feature analysis algorithm to ensure certain accuracy and real-time performance. On the basis, three-dimensional dynamic gestures with a certain degree of discrimination, such as moving, rotating, ticking, crossing, shaking, and tapping, are recognized respectively, so as to realize corresponding applications. However, since Patent Document 1 is only for single-handed gesture recognition, there is an urgent need for a two-handed three-dimensional dynamic gesture recognition method.

Contents of the invention

The object of the present invention is to overcome the problems of inaccurate and inconvenient recognition of two-hand gestures, and provide a method and system for two-hand three-dimensional dynamic gesture recognition.

For reaching above-mentioned object, the present invention realizes by following technical scheme:

A two-hands three-dimensional dynamic gesture recognition method, using two mobile terminals, the two mobile terminal devices adopt a master-slave mode, and one of the mobile terminals is used as the master, which is called the master mobile terminal, and the other mobile terminal is used as the slave. It is called the slave mobile terminal; the two mobile terminals track the gestures of the left hand and the right hand respectively; the master mobile terminal scans the external devices that can be connected to the outside world, and establishes a wireless data connection with the slave mobile terminal; the master mobile terminal collects gesture data; the slave mobile terminal Collect gesture data, and send the gesture data to the main mobile terminal in real time through the wireless data connection for data processing, synchronization and two-hand gesture recognition; the main mobile terminal sends the recognition result of the main mobile terminal and the gesture data obtained from the mobile terminal to the The device to be controlled.

Further, the mobile terminal device is integrated with a 9-axis sensor module, wherein a 3-axis accelerometer measures acceleration, a 3-axis gyroscope measures angular velocity, and a 3-axis magnetometer measures magnetic field.

Further, the synchronization process includes: using the key value detection KVO of the mobile terminal operating system to monitor whether the two mobile terminal devices have completed the collection and recognition of gestures. If one of them is completed, it enters a waiting state, and external operations are invalid for it; until When the other party also completes the collection and identification process, it packs the data instructions and sends them to the device to be controlled, and then makes the two mobile terminal devices enter the ready state, waiting for the next gesture with both hands.

Further, when the gesture is intercepted, a stronger threshold Ats is used to detect the movement of the gesture, and on this basis, two smaller thresholds Atb and Atf are used to judge the start and end of the gesture, where Ats >Atf, Ats>Atb; it is necessary to limit the time difference between Ats and Atf, Atb and the two, so as to avoid dividing the continuous action of a gesture into multiple gestures; in addition, it is necessary to limit the time difference between Atf and Atb to avoid the user’s hand Part of the occasional jitter is recognized as the beginning of the gesture; in order to ensure the integrity of the gesture data, the time series data intercepted at Atb and Atf need to be appropriately extended to both ends of the time length Te.

Specifically, first detect the time series points of Ats, and then search forward and backward on the basis of Ats to find the time series points of Atf and Atb. On top of this, appropriately expand the length of time Te to obtain the Complete acceleration data.

On the other hand, the present invention also proposes a two-hands three-dimensional dynamic gesture recognition system, the system includes two mobile terminals, and a device to be controlled; each of the mobile terminals includes a gesture data acquisition unit, a wireless data communication Unit, two mobile terminal devices adopt the master-slave mode, one of the mobile terminals is used as the master, which is called the master mobile terminal, and the other mobile terminal is used as the slave, which is called the slave mobile terminal; the two mobile terminals track respectively Gestures of the left hand and right hand; the master mobile terminal scans external devices that can be connected to the outside world, and establishes a wireless data connection with the slave mobile terminal; the gesture data collection unit of the master mobile terminal collects gesture data; the gesture data collection unit of the slave mobile terminal collects gesture data , and send the gesture data to the main mobile terminal in real time through the wireless data communication unit for data processing, synchronization and gesture recognition with both hands; The result is sent to the device to be controlled.

Further, the wireless data communication unit is a Bluetooth module, a WiFi module or an RF module.

Further, the device to be controlled is a PC, a game console, a drone or a VR helmet.

Further, the master mobile terminal is a smart phone, and the slave mobile terminal is a wearable device. Alternatively, both the master mobile terminal and the slave mobile terminal are smart phones. Alternatively, both the master mobile terminal and the slave mobile terminal are wearable devices.

The beneficial effects of the present invention are: the two-handed gesture recognition method of the present invention directly uses human hands as the input terminal to the smart device, and the interaction between man and machine does not need other media, which improves the efficiency of interaction and avoids the problems caused by intermediate media. Various misidentifications; expand the application occasions of smart devices, so that human-computer interaction is no longer limited to specific parts of the device, allowing users to perform certain operations when it is inconvenient to speak or take out the device, such as Use certain gestures to answer mobile phone calls while washing clothes; enrich the application scenarios of smart devices, for example, the combination of two-hand gesture operation and virtual scene can enrich the operability and entertainment of handheld devices; two-hand gesture input provides a A new idea of robot demonstration learning, an important part of gesture input is user-defined gestures; the 9-axis sensor in the IOS smart mobile terminal effectively increases the accuracy of recognition and enriches the realized functions; it adopts wireless communication technologies such as Bluetooth For data transmission, low power consumption, and supports simultaneous use by multiple users.

Description of drawings

Fig. 1 is the hardware block diagram of the double gesture recognition system that the method of the present invention is based on;

Fig. 2 is a schematic diagram of hand gesture data synchronization;

Fig. 3 is the flowchart of the implementation scheme of the gesture recognition system of the mobile device;

Figure 4(a) is a schematic diagram of gesture data interception based on FBGD;

Figure 4(b) is the intercepted data based on FBGD gesture data.

specific implementation plan

The present invention will be further described in detail below through specific embodiments in conjunction with the accompanying drawings.

Smart mobile devices have been fully popularized, and the development of gesture recognition based on mobile terminal platforms will allow more people to use the convenience of gesture recognition. The mobile terminal device integrates a 9-axis sensor module (3-axis accelerometer to measure acceleration, 3-axis gyroscope to measure angular velocity, and 3-axis magnetometer to measure magnetic field), and has high recognition accuracy, which can realize effective recognition of complex gestures. The present invention utilizes the cooperation and interaction of the mobile phone and the watch or bracelet to realize data interaction and gesture interaction with both hands, which greatly enriches the function of a single gesture.

The hardware block diagram of the dual-gesture recognition system based on the method of the present invention is shown in Figure 1, including two mobile terminal devices and devices to be controlled, such as PCs, game consoles, and drones. The communication between the three parties is realized through Bluetooth technology. The two mobile terminal devices adopt the master-slave mode, and one of the mobile terminals (such as iPhone) is used as the master, which is called the master mobile terminal, and the other mobile terminal (such as iWatch) is used as the slave, which is called the slave mobile terminal; The two mobile terminals track the gestures of the left hand and the right hand respectively (for example, the left hand holds the iPhone and the right hand wears the iWatch). The master mobile terminal scans the external devices that can be connected to the outside world, establishes a Bluetooth data connection with the slave mobile terminal; collects gesture data from the mobile terminal, and sends the gesture data to the master mobile terminal in real time through Bluetooth for data processing, synchronization and two-hand gesture recognition ; Next, send the recognition result of the master mobile terminal and the gesture data obtained from the mobile terminal to the Bluetooth receiving module of the device to be controlled; finally, the device to be controlled performs corresponding control operations according to the recognition result.

Since there is a sequence of actions in the gestures of both hands, when one party collects gestures, the other party may not have started or is still in the process of recognition. This has caused the problem of asynchronous data of both hands, so a set of data completion protocol must be developed to control the asynchronous problem of gestures. The present invention proposes to use the key-value detection (KVO) of the operating system of the mobile terminal to monitor whether the two devices have completed the collection and recognition of gestures. If one party completes, it enters the waiting state, and external operations are invalid for it. When the other party also completes the collection and identification process, the data command is packaged and sent to the device to be controlled, and then the device enters the ready state, waiting for the next gesture of both hands. Its operating principle is shown in Figure 2.

The essence of gesture recognition is to use gesture recognition algorithms to classify gesture actions according to the gesture model. The pros and cons of gesture recognition methods are directly related to the efficiency and accuracy of gesture recognition. Common gesture recognition methods include:

(1) The DTW algorithm is a nonlinear warping technique that combines time warping and distance measure calculation, and has a nonlinear time normalization effect. Using a nonlinear regularization function with specified properties to approximate the fluctuations on the time axis, by stretching the time axis of one of the modes to maximize the overlap with the other mode, the residual distance is minimized, thereby eliminating the two Spatiotemporal representations of time differences between patterns. In fact, it is a simplification of the hidden Markov model. For simpler time series, the two of them are equivalent. Methods allow sufficient flexibility between test and reference patterns to enable classification.

(2) Hidden Markov model is an extension of Markov model. The Markov model describes a random process and transitions between states. Hidden Markov model describes two random processes, one random process describes the probability relationship between the output and the state, that is, the output is a random process function of the state, and the other stochastic process describes the transition relationship between the states. The observer can see the output, but cannot see the transition between states, that is, the transition between states is implicit. Due to the particularity of the hidden Markov model topology, it is too complicated when analyzing gesture signals, which makes the calculation of training and recognition large, especially in the continuous hidden Markov model, it is necessary to calculate a large number of state probability densities , the number of parameters to be estimated is large, making the speed of training and recognition relatively slow.

(3) Artificial neural network. In the field of gesture recognition, artificial neural network is a very widely used tool. Artificial neural network has self-organization and self-learning ability, strong anti-noise ability, strong fault tolerance and robustness. The artificial neural network forms a complex information processing network by extensively connecting a large number of simple processing units. The processing units and their interconnection modes are designed by referring to the structure and connection mechanism of human brain neurons. This kind of network has the ability of learning and memory, knowledge generalization and input information feature extraction similar to the human brain. After years of development, artificial neural networks already have many models, such as fuzzy neural network and BP neural network. The BP neural network based on the backpropagation learning algorithm is widely used at present.

(4) Machine learning, common classification methods based on machine learning include K_nearest neighbor method and support vector machine method. The basic idea of the K_nearest neighbor method is to formalize the text content into a weighted feature vector in the feature space according to the traditional vector space model. For a test object, calculate its similarity with each sample in the training sample set, and find out K most similar texts, according to the weighted distance to judge the category of the test object. The support vector machine takes the training error as the constraint condition of the optimization problem, and takes the minimization of the confidence range value as the goal. It is a learning method based on the criterion of structural risk minimization. The computational complexity of SVM depends on the number of support vectors, not the dimensionality of the sample space, which in a sense avoids the "dimensional space". Moreover, in this method, adding and deleting non-support vector samples has no effect on the model, and it is not very sensitive to the choice of function. In addition, the support vector sample set has a certain robustness.

In addition, there are methods of combining HMM and threshold comparison, methods of combining HMM and neural network, methods of combining Bayesian network and support vector machine, and recognition methods using features such as gesture tilt characteristics.

Because the speed of gesture recognition is high, the present invention adopts the method of combining the E_DTW dynamic time template matching method and feature rough classification. Therefore, when selecting a gesture action, it is necessary to observe and analyze a large number of samples of the gesture. Through sample collection and waveform observation, the standard template for gesture recognition is obtained, and a custom gesture recognition database is constructed to realize dynamic gesture template matching. Currently, the following 8 types of gestures are temporarily used as the default gestures of the system. As shown in Table 1:

Table 1 Gesture set definition

The above-mentioned gestures have a certain degree of discrimination, and the feature quantities formed by their respective sensor information have relatively high recognizability, so different classifiers can be established according to the feature values of different gestures. First, the eight categories of gestures are pre-divided into 3 categories according to the calculated eigenvalues, and the categories are identified, and then the algorithm is designed to subdivide the rotation and movement movements, and identify the direction of motion, so that we want to use subdivision The eigenvalue analysis method realizes the classification and recognition of gestures quickly and efficiently.

The process flow of the gesture recognition system implementation scheme for mobile devices is shown in Figure 3.

DTW template matching recognition technology, its core idea is to match the input raw data with the pre-stored template, and complete the recognition task by measuring the similarity between the two templates. Commonly used distance similarity calculation methods include weighted Euclidean distance method, correlation coefficient method and logarithmic distance method. Template matching needs to solve the problem that the input data is inconsistent with the length of the pre-stored template time series, because even for the same gesture, its duration will change randomly. To solve this time alignment problem, a typical timing template matching is DTW.

Pre-classify the length characteristics of gesture data, and realize the pre-classification of gestures with large differences according to the length and energy of different gestures. Therefore, the premise is to effectively intercept the gesture data, that is, intercept the paragraph from the start point to the end point of the gesture from all the read raw data, and then convert the coordinate system of the intercepted data segment. The gesture data segment after cutting and coordinate system conversion can be used to extract features for gesture recognition. Next, the gesture cutting, coordinate system conversion and feature recognition methods will be introduced in detail.

Due to the inevitable shaking of the hand during the user's movement or gestures and the influence of the accuracy of the sensor itself, the gesture data collected by the acceleration sensor is inevitably disturbed by noise, resulting in up and down jitter. Smooth denoising is mainly to process the interference data such as environmental noise, and minimize the impact as much as possible. I use a simple moving average filter method to smooth and denoise the acquired acceleration data. SMA can filter out random noise while maintaining fast response. The derivation formula of the SMA filter is as follows:

SMA _now ＝(X _i +X _i-1 +....+X _i-n+1 )/n; n＝1,2,3,4,....

In the above formula, n represents the length of the data sequence. The size of n is related to the smoothing effect. If n is too small, the stabilization effect is not obvious. If n is too large, the stabilization effect is good, but it is easy to cause gesture information loss. Referring to experience, depending on different situations, n generally takes 5 to 15.

When calculating, you can directly use the following formula

SMA _now = SMA _previous -X _in /n+X _i /n.

The determination of the start and end of a gesture action plays a crucial role in gesture recognition. The previous part of this article has mentioned that most gesture recognition based on acceleration sensors require additional operations such as buttons to inform the gesture recognition system of the start and end of the gesture during the gesture execution process. By pressing the button to determine the start and end of the gesture, it is indeed possible to ensure that the gesture recognition system accurately samples the gesture action, but it has the following shortcomings. Gesture data brings noise interference. The second is that the user needs to determine the position of the button through the eyes when pressing the button, which distracts the user's attention, reduces the user experience, and is not conducive to a more free and natural human-computer interaction. Especially for blind people or people with inflexible fingers, it is more difficult to press buttons and other operations, which is not conducive to the promotion and popularization of gesture recognition systems.

In threshold-based gesture judgment, the setting of the threshold largely determines the accuracy of the judgment. If the threshold is too low, the user's inadvertent subtle movements will be judged as the beginning of the gesture, causing misjudgment. If the threshold is too high, the judgment of normal gestures will be missed. Therefore, in combination with the characteristics of gesture actions, a gesture determination method using FBGD is invented. A strong threshold Ats is used to detect the motion of the gesture, and on this basis, two smaller thresholds Atf and Atb are used to judge the start and end of the gesture. Ats, Atf and Atb are within the time T. Threshold for the difference between the maximum and minimum values of the acceleration data series. At the same time, we limit the size of the time difference between Ats and Atf, Atb and the two, so as to avoid dividing the continuous action of one gesture into multiple gestures. In addition, it is necessary to limit the size of the time difference between Atf and Atb, so as to avoid recognizing the occasional shaking of the user's hand as the beginning of the gesture. In order to ensure the integrity of the gesture data, the time series data intercepted at Atb and Atf need to be appropriately extended to both ends of the time length Te. As shown in Figure 4(a) and Figure 4(b), when the gesture data is smoothed and sent to gesture detection, the time series point of Ats is first detected, and then based on Ats, it passes forward , and search backward to find the time series points of Atf and Atb, above which, the length of time Te is appropriately extended to obtain the complete acceleration data of the gesture.

Different gesture data have different data lengths and different amplitudes. Even among the same gestures of the same person, there are differences. Therefore, the data must be normalized to the same amplitude range and sampled to the same length. The advantages of this are: 1. Unify the gesture quantification standard to avoid excessive or excessive amplitude. Small gestures are misjudged; 2. On the premise of ensuring that the waveform is not distorted, the fixed sampling point actually reduces the computational complexity and improves the recognition speed.

The gesture recognition algorithm used in the present invention includes but is not limited to the above-mentioned algorithm, and the fast three-dimensional dynamic gesture recognition method based on the subdivision eigenvalue method described in Patent Document 1 can also be used.

Example 1

Traditional UAV flight control generally uses RF radio frequency remote control. It has a long antenna outside, and the remote control commands are all through the control switches and buttons outside the casing, modulated and coded by the internal circuit, and then the electromagnetic waves are emitted by the antenna through the high-frequency signal amplification circuit. There are three types of remote control transmitters commonly used at present: box-type button-type handheld, portable rod-type remote control, and hand-held gun-type remote control. For example, the remote control stick transmitter has two joysticks, the left stick is used to control the ascent and descent of the drone, and the right stick controls the flight direction of the drone. The center is equipped with a liquid crystal display screen to display the working status and functions. Its advantages are long communication distance and high functional integration, which can adapt to the flight control of UAVs in different environments. However, it has high requirements for personnel operation, and generally requires professional training.

With the in-depth study of gesture recognition, gesture recognition can be combined with UAV control to realize gesture control of UAV flight trajectory. The space gesture trajectory is recognized through the mobile terminal device, and the recognition result is sent to the drone through the wireless device to control the command. Under the premise of ensuring the stable and safe flight of the UAV, users can customize the flight trajectory of the UAV. The invention is based on the dynamic gesture trajectory recognition of the MEMS sensor. Train the valid trajectory samples and recognize the centroid trajectory of the input gesture, so as to meet the needs of controlling the flight trajectory of the drone. The gesture is integrated with the aircraft system, and the gesture control of the drone's flight trajectory is realized in the system.

Example 2

In the first-person perspective game, combined with gesture recognition, the gesture recognition device platform is used as a gamepad. In addition, the currently popular somatosensory games based on machine vision on the market are still not sensitive enough, and the experience effect is not good. If the camera can be integrated on the smart terminal and combined with the vision and hardware platform, the position of the hardware platform can be determined, and then fine positioning can be carried out through the gyroscope and other data on the hardware platform, which greatly improves the recognition accuracy and sensitivity. Due to the recognition of both hands, it can be used in many large-scale competitive fighting games to provide a better human-computer interaction experience.

Example 3

Two-handed devices, combined with VR helmets, can realize immersive user gaming experience, use the helmets to project 3D images, and combine with mobile devices, you can realize immersive gaming experience. Because of the expansibility of both hands, it can realize many interactive modes, and can provide users with rich functions that ordinary operating handles cannot provide.

Example 4

There are many operations that require directional control on the PC side, such as video playback, picture switching, and web page sliding up and down. Traditional mouse control has a relatively large distance limit and low maneuverability, which affects the friendly interaction between people and the PC side.

At present, fast forward and rewind when watching video on the PC end are basically realized through the mouse and keyboard, but it is not very convenient for users. Therefore, it is proposed to use a portable mobile device to use gesture recognition to control video playback. , or the zooming and switching functions of the picture viewer. Four kinds of gestures can be defined, which correspond to these four kinds of commands respectively. Finally, the purpose of controlling the playback of the player is achieved through gestures, and due to the popularity of gesture recognition devices, it further satisfies most users.

The above content is a further detailed description of the present invention in conjunction with specific preferred embodiments, and it cannot be assumed that the specific implementation of the present invention is limited to these descriptions. For those of ordinary skill in the technical field of the present invention, without departing from the concept of the present invention, some simple deduction or replacement can be made, which should be regarded as belonging to the protection scope of the present invention.

Claims (10)

1. A two-hands three-dimensional dynamic gesture recognition method is characterized in that: the method utilizes two mobile terminals, and the two mobile terminal devices adopt a master-slave mode, and one of the mobile terminals is used as a host, which is referred to as a master mobile terminal, The other mobile terminal acts as a slave, called the slave mobile terminal; the two mobile terminals track the gestures of the left hand and the right hand respectively; the master mobile terminal scans for external devices that can be connected to the outside world, and establishes a wireless data connection with the slave mobile terminal; The mobile terminal collects gesture data; collects gesture data from the mobile terminal, and sends the gesture data to the main mobile terminal in real time through a wireless data connection for data processing, synchronization and two-hand gesture recognition; the main mobile terminal combines the main mobile terminal with the slave mobile terminal. The result recognized by the gesture data is sent to the device to be controlled. 2. The method according to claim 1, wherein a 9-axis sensor module is integrated on the mobile terminal device, wherein a 3-axis accelerometer measures acceleration, a 3-axis gyroscope measures angular velocity, and a 3-axis magnetometer measures a magnetic field . 3. The method according to claim 1, characterized in that: the synchronization process includes: using the key value detection KVO of the mobile terminal operating system to monitor whether the two mobile terminal devices have completed the collection and recognition of gestures, if one party completes , enter the waiting state, and the external operation is invalid for it; until the other party also completes the collection and identification process, the data command is packaged and sent to the device to be controlled, and then the two mobile terminal devices enter the ready state, waiting for the next two-hand gesture. 4. The method according to claim 1, characterized in that: when the gesture is intercepted, a stronger threshold Ats is used to detect the motion of the gesture, and on this basis, two smaller thresholds Atb, Atf is used to judge the start and end of the gesture, among them, Ats>Atf, Ats>Atb; it is necessary to limit the time difference between Ats and Atf, Atb and the two, so as to avoid dividing the continuous action of a gesture into multiple gestures; in addition, it is necessary Limit the time difference between Atf and Atb to avoid recognizing the occasional shaking of the user's hand as the beginning of the gesture; in order to ensure the integrity of the gesture data, the time series data intercepted at Atb and Atf need to be appropriately extended to both ends of the time length Te. 5. The method according to claim 4, characterized in that: specifically, at first detect the time series point of Ats, then on the basis of Ats by searching forward and backward, find the time series point of Atf and Atb , on top of this, the length of the time Te is appropriately extended to obtain the complete acceleration data of the gesture. 6. The method according to claim 1, wherein the wireless data connection is Bluetooth, Wifi or RF. 7. A three-dimensional dynamic gesture recognition system for both hands, characterized in that: the system includes two mobile terminals and a device to be controlled; the two mobile terminal devices adopt a master-slave mode, and one of the mobile terminals is used as a host, called The main mobile terminal is the master mobile terminal, and the other mobile terminal is the slave machine, which is called the slave mobile terminal; the two mobile terminals track the gestures of the left hand and the right hand respectively; the master mobile terminal scans the external devices that can be connected to the outside world, and establishes a Wireless data connection; the main mobile terminal collects gesture data; collects gesture data from the mobile terminal, and sends the gesture data to the main mobile terminal in real time through the wireless data connection for data processing, synchronization and two-hand gesture recognition; the main mobile terminal transfers the main mobile terminal The result identified with the gesture data obtained from the mobile terminal is sent to the device to be controlled. 8. The system according to claim 7, wherein a 9-axis sensor module is integrated on the mobile terminal device, wherein a 3-axis accelerometer measures acceleration, a 3-axis gyroscope measures angular velocity, and a 3-axis magnetometer measures magnetic field . 9. The system according to claim 7, characterized in that: the synchronization process includes: using the key value detection KVO of the mobile terminal operating system to monitor whether the two mobile terminal devices have completed the collection and recognition of gestures, if one party completes , enter the waiting state, and the external operation is invalid for it; until the other party also completes the collection and identification process, the data command is packaged and sent to the device to be controlled, and then the two mobile terminal devices enter the ready state, waiting for the next two-hand gesture. 10. The system according to claim 7, wherein the master mobile terminal is a smart phone, and the slave mobile terminal is a wearable device.