Intelligent interaction system applied to AI glasses

Abstract

The invention discloses an intelligent interaction system applied to AI glasses, relates to the technical field of man-machine interaction, and is used for solving the problems that resource utilization redundancy is increased and equipment power consumption efficiency is reduced. Fusing to form a head posture variable quantity, and analyzing the dispersion degree of the head posture variable quantity based on an MAD algorithm to judge the complexity of an interaction scene; selecting to execute resource optimization or enhance an interaction response strategy according to the complexity, such as improving the speech recognition accuracy and adapting to a complex sound field; further introducing a proportionality coefficient of illumination intensity, movement speed and noise intensity, constructing a long-short-term memory neural network model to predict interaction complexity, realizing dynamic interaction mode adjustment through a gating mechanism and a regularization method, and judging whether to improve the interaction mode according to prediction output. And the self-adaption and high-reliability interaction capability of the AI glasses in a complex environment is ensured.

Description

Technical Field

[0001] The present invention relates to the field of human-computer interaction technology. More specifically, the present invention relates to an intelligent interaction system applied to AI glasses. Background Art

[0002] With the rapid development of artificial intelligence, wearable devices, and human-computer interaction technology, AI glasses, as intelligent terminals integrating display, perception, and computing functions, have gradually been widely deployed in application scenarios such as navigation assistance, real-time translation, and industrial remote guidance. Existing AI glasses usually integrate image sensors, speech recognition modules, and inertial measurement units, and achieve human-computer interaction through multi-modal methods such as touch, voice, and gesture.

[0003] The existing technology has the following deficiencies: Currently, most of the human-computer interaction methods in AI glasses rely on fixed gesture recognition, voice command triggering, and preset interaction processes, lacking the ability to real-time perceive natural actions such as head postures, unable to dynamically recognize fine-grained changes in user intentions, the interaction strategy lacks scene adaptability, and the system cannot dynamically allocate resources according to the interaction complexity, resulting in an increase in redundant resource utilization and a decrease in device power consumption efficiency. Therefore, an intelligent interaction system applied to AI glasses is proposed.

[0004] The above information disclosed in the background art section is only used to enhance the understanding of the background of the present disclosure. Therefore, it may include information that does not constitute the prior art known to those of ordinary skill in the art. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the existing technology, the embodiments of the present invention provide an intelligent interaction system applied to AI glasses, which uses an intention recognition mechanism based on head posture changes, a dynamic evaluation algorithm for interaction scene complexity, a resource scheduling optimization strategy, and an interaction mode adaptive regulation mechanism to solve the problems of lagging interaction response, low resource utilization rate, and non-dynamically optimizable mode proposed in the above background art, so as to achieve intelligent response, dynamic resource configuration, and refined interaction experience of AI glasses during the human-computer interaction process.

[0006] To achieve the above object, the present invention provides the following technical solution. An intelligent interaction system applied to AI glasses includes a data acquisition module, an intention recognition module, a strategy decision module, and a mode optimization module. The functions of each module are as follows: The data acquisition module collects head posture data through an inertial measurement unit integrated inside the AI glasses, and calculates the head posture change amount according to the head posture data; The intention recognition module analyzes the head posture change amount and judges the complexity of the interaction scene; The policy decision module determines whether to adjust the interaction policy according to the complexity of the interaction scenario. If the complexity of the interaction scenario is low, system resource optimization is performed; if the complexity of the interaction scenario is high, the interaction response ability is enhanced, and the resource optimization result is evaluated based on the interaction response ability. The mode optimization module determines whether to improve the interaction mode according to the system resource optimization result.

[0007] In a preferred embodiment, an X, Y, Z three-axis coordinate system is constructed with the geometric center of the inertial measurement unit integrated inside the AI glasses as the origin, and the head pose data is the linear acceleration and angular velocity along the X, Y, Z three-axis coordinate system measured by the inertial measurement unit. Taking the current moment as the starting point, a fixed time period is selected forward as the acquisition period, the angular velocities of the X, Y, Z three-axis coordinate system are acquired, and the modulus length of the angular velocities is calculated to obtain the rotation synthesis value. Select the same fixed time period before the acquisition period as the previous acquisition period, respectively acquire the linear accelerations in the previous sampling period and the acquisition period and take the difference, and then calculate the modulus length of the difference to obtain the acceleration change intensity.

[0008] In a preferred embodiment, a fixed number of the same fixed time periods before the acquisition period are selected. Obtain the maximum rotation synthesis value and acceleration change intensity within all the same fixed time periods. Divide the rotation synthesis value and acceleration change intensity within the acquisition period by the maximum rotation synthesis value and acceleration change intensity within all the same fixed time periods respectively to obtain the normalized values. Fuse the normalized values according to the preset weighting coefficients to obtain the head pose change amount.

[0009] In a preferred embodiment, the position coordinates of the head are obtained, and a spherical area with a fixed diameter is extended around the position coordinates of the head in all directions in space as the sample range. Obtain all the head pose change amounts within the sample range, and merge the head pose change amounts into a change amount data set. Arrange the head pose change amount data in the change amount data set in ascending order, and obtain the median of the head pose change amounts in the change amount data set. Calculate the absolute deviation of each head pose change amount from the median to form an absolute deviation set. Take the median of the absolute deviation set to obtain the MAD value.

[0010] In a preferred embodiment, the MAD value is compared with the interaction scenario complexity judgment threshold. When the MAD value is greater than the interaction scenario complexity judgment threshold, it is determined that the complexity of the interaction scenario is high. When the MAD value is less than or equal to the interaction scenario complexity judgment threshold, it is determined that the interaction scenario complexity is low.

[0011] In a preferred embodiment, if the interaction scenario complexity is low, system resource optimization is directly performed; If the interaction scenario complexity is high, after improving the interaction response ability, system resource optimization is performed; When the interaction scenario complexity is low, the angular velocity and acceleration of the head pose are obtained, and in combination with the operating frequency of the system, the response is adjusted until the head pose change amount reaches the head pose change amount threshold to achieve resource optimization.

[0012] In a preferred embodiment, when the interaction scenario complexity is high, the speech recognition accuracy, noise intensity, and ambient light intensity within the sample range are obtained to comprehensively judge the interaction response ability; If the predicted speech recognition accuracy is not within the preset speech recognition accuracy interval, the speech recognition accuracy is improved through an adaptive filtering algorithm; The ambient light intensities are combined into a light intensity data set, and the median of the ambient light intensities is taken as the light intensity threshold; The ambient light intensities are successively divided by the light intensity threshold to obtain the respective light intensity coefficients.

[0013] In a preferred embodiment, when the light intensity coefficient is within the preset light intensity coefficient interval threshold, the corresponding ambient light intensity is retained and the average value is taken to obtain the average light intensity coefficient; The ratio of the predicted light intensity of the interaction scenario to the average light intensity coefficient is used as the light intensity ratio coefficient; The motion speed data within the sample range, the angular velocity of the predicted interaction scenario, and the angular velocity values within the sample range are obtained, and the motion speed ratio coefficient is comprehensively calculated; The ratio of the noise intensity to the preset noise intensity threshold is used as the noise intensity ratio coefficient.

[0014] In a preferred embodiment, a long short-term memory network model is constructed through the light intensity ratio coefficient, the motion speed ratio coefficient, and the noise intensity ratio coefficient to obtain the predicted output value; If the predicted output value is less than the preset output threshold, the interaction mode is not improved; If the predicted output value is greater than the preset output threshold, the interaction mode is improved.

[0015] The technical effects and advantages of the present invention: The present invention collects triaxial acceleration and angular velocity data through an inertial measurement unit integrated in AI glasses, calculates the rotation synthesis value and the acceleration change intensity, fuses them into the head posture change amount, analyzes the dispersion degree of the change amount based on the MAD algorithm, thereby judging the complexity of the interaction scenario, and selects to execute resource optimization or enhance the interaction response strategy according to the scenario complexity, such as improving the speech recognition accuracy and adapting to complex sound fields. Further, the proportional coefficients of light intensity, motion speed, and noise intensity are introduced to construct a long short-term memory neural network model to predict the interaction complexity, and the gating mechanism and regularization method are used to achieve efficient and dynamic adjustment of the interaction mode. Whether to improve the interaction mode is judged according to the predicted output value, ensuring that the AI glasses have adaptive and highly reliable intelligent interaction capabilities in different complex environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a schematic diagram of the steps of the intelligent interaction system of the present invention applied to AI glasses.

[0017] Figure 2 It is a flowchart of the implementation of the intelligent interaction system of the present invention applied to AI glasses. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0018] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0019] Embodiment 1, an intelligent interaction system applied to AI glasses, as Figure 1 and Figure 2 shown, includes a data acquisition module, an intention recognition module, a strategy decision module, and a mode optimization module. Each module is electrically connected, and the functions are as follows: The data acquisition module collects head posture data through an inertial measurement unit integrated inside the AI glasses, and calculates the head posture change amount according to the head posture data; The intention recognition module analyzes the head posture change amount using the MAD algorithm, and judges the complexity of the interaction scenario accordingly; The strategy decision module judges whether it is necessary to adjust the interaction strategy according to the complexity of the interaction scenario. If the interaction scenario complexity is low, system resource optimization is performed; if the interaction scenario complexity is high, the interaction response ability is enhanced, and the resource optimization result is evaluated based on the interaction response ability; The mode optimization module judges whether it is necessary to improve the interaction mode according to the system resource optimization result.

[0020] The specific implementation is as follows: In the data acquisition module, head pose data generated during the user's use of the AI glasses is acquired. The AI glasses are integrated with an inertial measurement unit, which consists of a three-axis accelerometer and a three-axis gyroscope. The head pose data is the linear acceleration and angular velocity measured by the three-axis accelerometer and the three-axis gyroscope respectively along the X, Y, and Z coordinate axes. The origin of the X, Y, and Z coordinate axes is located at the geometric center of the inertial measurement unit, and this geometric center is fixed on the upper edge of the nose pad of the glasses frame, maintaining a rigid connection with the user's head movement.

[0021] Select a fixed time period forward from the current moment as the acquisition period, acquire the angular velocity of the X, Y, and Z coordinate axes, calculate the magnitude of the angular velocity to obtain the rotation synthesis value, which reflects the intensity of the head's rotational movement during the acquisition period. The specific calculation formula is ; Among them, is the angular velocity of the X, Y, and Z axes, is the rotation synthesis value, which is used to describe the comprehensive degree of the head pose changing around any axis. The larger the value, the more significant the rotation amplitude.

[0022] Select the same fixed time period before the acquisition period as the previous acquisition period, record the linear accelerations within the previous sampling period and the acquisition period respectively, subtract the linear acceleration within the acquisition period from the linear acceleration within the previous sampling period, and calculate the magnitude of the difference, thereby obtaining the acceleration change intensity, which reflects whether the head makes a sudden movement. The specific calculation formula is as follows: ; Among them, is the linear acceleration of the X, Y, and Z axes within the acquisition period, is the linear acceleration of the X, Y, and Z axes within the previous acquisition period, is the acceleration change intensity. The larger the value, the more significant the acceleration change.

[0023] To unify the dimension and ensure data comparability, select a fixed number of the same fixed time periods before the acquisition period, calculate the rotation synthesis value and the acceleration change intensity within all fixed time periods, divide the rotation synthesis value and the acceleration change intensity within the acquisition period by the maximum rotation synthesis value and acceleration change intensity within all the same fixed time periods respectively to obtain the normalized values. The range of the normalized values is limited to interval, and fuse the normalized rotation change intensity and acceleration change intensity according to the preset weighting coefficient to obtain the head pose change amount. The specific calculation formula is: ; Among them, is the head pose change amount, is a preset weight parameter that satisfies , are respectively the maximum values of the rotation synthesis value and the acceleration change intensity within all fixed time periods.

[0024] It should be noted that the inertial measurement unit is a sensor unit integrated inside the AI glasses, mainly used to collect the position coordinates and dynamic motion information of the head in real time. It consists of a three-axis accelerometer and a three-axis gyroscope, which are respectively used to measure the linear acceleration and angular velocity of the head along three orthogonal axes in space; its modulus length is the square root of the sum of the squares of the components of any three-dimensional vector, used to measure the magnitude of the vector, which will not be elaborated here.

[0025] In the intention recognition module, the position coordinates of the head are obtained through the inertial measurement unit. A spherical area with a fixed diameter is extended around the position coordinates of the head in all directions in space as the sample range, and all head pose change amounts within the sample range are obtained. The head pose change amounts are combined into a change amount data set, and the MAD algorithm is used to analyze the change amount data set to determine the complexity of the current interaction scenario of the user; The head pose change amount data in the change amount data set is arranged in ascending order, the median of the head pose change amounts in the change amount data set is obtained, and for each head pose change amount, its absolute deviation from the median is calculated to form an absolute deviation set; The median of the absolute deviation set is taken to obtain the MAD value. This MAD value is used to measure the overall dispersion degree of the head pose change amounts, that is, the fluctuation amplitude of the user's head movements. The larger the MAD value, the more frequent the head movement changes and the less concentrated the pose change amounts, indicating that the user is in a more active and frequent interaction state at the current stage and the complexity of the interaction scenario is higher; on the contrary, if the MAD value is smaller, it means that the user's action changes are stable or weaker, the interaction intention is not obvious, and the complexity of the interaction scenario is lower.

[0026] The MAD value is compared with the interaction scenario complexity judgment threshold. When the MAD value is greater than the interaction scenario complexity judgment threshold, it is determined that the complexity of the interaction scenario is high; When the MAD value is less than or equal to the interaction scenario complexity judgment threshold, it is determined that the complexity of the interaction scenario is low; It should be noted that the MAD algorithm is a robust statistical method used to measure the dispersion degree of a set of data; the interaction scenario complexity judgment threshold is a preset value obtained by professionals based on a large number of historical sample experiments, representing the judgment baseline of the system for the activity degree of interaction behavior, which will not be elaborated here.

[0027] In the policy decision-making module, if the complexity of the interaction scenario is low, system resource optimization is directly performed; if the complexity of the interaction scenario is high, the interaction response ability is enhanced before performing system resource optimization.

[0028] When the complexity of the interaction scenario is low, the angular velocity and acceleration of the head pose are further obtained through an inertial measurement unit, and response adjustment is performed in combination with the operating frequency of the system. When the change amount of the head pose reaches the threshold of the head pose change amount after dynamically adjusting the operating frequency, it can be confirmed that the current interaction effect meets the resource optimization conditions, and then the resource optimization operation is completed.

[0029] Among them, the operating frequency of the system is the execution rate at which the core processing module responsible for the interaction response task in the AI glasses completes the data processing task per unit time.

[0030] When the complexity of the interaction scenario is high, the speech recognition accuracy and noise intensity are detected through the speech signal processing module, and the ambient light intensity is collected through the photosensitive sensor built in the AI glasses.

[0031] The ambient light intensity usually does not directly affect the speech recognition accuracy, but can determine the stability of the user's visual focus, thus indirectly affecting the interaction response ability.

[0032] Among them, the speech signal processing module performs speech recognition through an array microphone, and takes the ratio of the speech recognition result to the speech command data as the speech recognition accuracy; after the microphone array collects the ambient background sound, the sound pressure level of the background sound is obtained through fast Fourier transform as the noise intensity.

[0033] If the predicted speech recognition accuracy is not within the preset speech recognition accuracy interval, the speech recognition accuracy is improved through an adaptive filtering algorithm. The adaptive filtering algorithm can adjust the weight coefficient or filter structure parameters of the filter in real time according to the current input speech command data and noise intensity, making the filter more suitable for the current acoustic environment characteristics, so as to suppress the interference noise to the greatest extent without losing the effective speech components.

[0034] The adaptive filtering algorithm identifies the frequency band with a low signal-to-noise ratio in the speech signal, and performs signal energy enhancement processing on this frequency band, or performs energy attenuation on the noise-dominated frequency band, so as to optimize the spectral structure of the speech signal and increase the proportion of the effective speech components in the overall signal.

[0035] The filtering output result is fed back to the speech recognition module in real time, and the parameters of the adaptive filter are iteratively updated according to the current speech recognition accuracy, so that the filtering strategy matches the speech recognition performance until the output recognition accuracy is stable within the preset speech recognition accuracy interval.

[0036] Through the above processing method, the voice input quality is automatically optimized in a complex noise environment, the dynamic adjustment of the voice recognition accuracy is realized, and the interactive system has good recognition performance and interactive reliability under different sound field conditions.

[0037] Collect the ambient light intensity through the photosensitive sensor built in the AI glasses, arrange the ambient light intensity within the sample range from small to large and merge them into a light intensity data set. Take the median of the light intensity data set as the light intensity threshold. Divide the light intensity in the light intensity data set by the light intensity threshold in turn to obtain the light intensity coefficient, and compare the light intensity coefficient with the preset light intensity coefficient interval threshold.

[0038] When the light intensity coefficient is within the preset light intensity coefficient interval threshold, retain the corresponding ambient light intensity; otherwise, remove the corresponding ambient light intensity.

[0039] Take the average of the retained light intensities to obtain the average light intensity coefficient, and use the ratio of the predicted light intensity of the interaction scene to the obtained average light intensity coefficient as the light intensity ratio coefficient.

[0040] The faster the user's movement speed and the more types of noise in the interaction scene, the lower the self-adjustment ability of the AI glasses system, resulting in a faster decline in the interaction response ability.

[0041] By accessing the motion monitoring website of the sample range location, obtain the motion speed data within the sample range as the reference benchmark for the interaction environment. The system collects the angular velocity of the predicted interaction scene and the angular velocity value of the sample range through the inertial measurement unit; divide the motion speed value within the sample range by the corresponding angular velocity value to obtain the motion speed density of the sample range.

[0042] Take the ratio of the angular velocity of the predicted interaction scene to the angular velocity value of the sample range as the angular velocity ratio coefficient. Multiply the motion speed density by the angular velocity ratio coefficient to obtain the motion speed density of the predicted interaction scene; Compare the motion speed density with the preset motion speed density threshold of the interaction scene to obtain the corresponding motion speed ratio coefficient.

[0043] The system collects the noise intensity in the predicted interaction scene through the voice signal processing module, and calculates the ratio of the noise intensity to the preset noise intensity threshold to obtain the noise intensity ratio coefficient.

[0044] The larger the motion speed ratio coefficient, the greater the motion speed density of the predicted interaction scene, the more types of noise in the predicted interaction scene, and the higher the interaction response ability; when the noise intensity ratio coefficient is larger, the noise intensity of the predicted interaction scene is greater, and the interaction response ability is higher.

[0045] Construct a long short - term memory network model through the light intensity proportionality coefficient, motion speed proportionality coefficient, and noise intensity proportionality coefficient to obtain the predicted output value.

[0046] Train the weight parameters of the hidden layer in the long short - term memory network model. During the training process, introduce the gated recurrent unit mechanism to achieve dynamic update of the memory cell state; at the same time, use the regularization method to compress the parameters of the output of the long short - term memory network model, and calculate the predicted output value; Compare the complexity prediction value with the preset output value to evaluate the complexity of the current interaction scenario, and judge whether it is necessary to optimize and adjust the interaction mode according to the complexity evaluation result. The specific implementation is as follows: Network state update: Update through the input data and the hidden state of the previous time step. This process can be represented by the following formula: , where, is the hidden state at time t, is the input data, is the input weight matrix, is the recurrent weight matrix, is the activation function.

[0047] Output layer calculation: Calculated through the hidden state and the output weight matrix. The output calculation formula is: , where, is the output value at time t, is the output weight matrix.

[0048] Training: The gradient descent method can be used to optimize the output weight matrix on the output layer. Take the light intensity proportionality coefficient, motion speed proportionality coefficient, and noise intensity proportionality coefficient as input data respectively, and sum the three outputs obtained through the above operations as the predicted output value.

[0049] Among them, the selection of time t, the input weight matrix, the weight recurrent matrix, and the output weight matrix involved above are set by professionals in this field according to the actual situation, and will not be analyzed here.

[0050] It should be noted that the long short-term memory network model is a variant of the neural network. In the weight matrix, the randomly selected weights are called memory states. The long short-term memory network model structure includes three main parts: an input gate, a forget gate, and an output gate. Among them, the input gate controls the inflow of information, the forget gate controls the forgetting of information, and the output gate controls the output of information. The gated recurrent unit is a simplified version of the long short-term memory network model used to describe the dynamic changes of memory units. The memory unit is assumed to be a dynamic system that can accumulate input signals and release output signals when certain conditions are met. The regularization method is a technique in machine learning used to prevent the model from overfitting. Overfitting means that the model performs well on the training data but poorly on the data that has not been run during testing. The photosensitive sensor is an electronic component that can convert the ambient light intensity into an electrical signal and is used in this example to collect the light intensity in real time. The adaptive filtering algorithm is a filtering technology with real-time self-adjusting capabilities that dynamically adjusts the structure or parameters of the filter according to the statistical characteristics of the input signal. The motion monitoring website at the location refers to the motion state data service platform in the area where the user is located, which can obtain real-time or statistical motion speed data in a specific area. All kinds of thresholds involved in the above steps are set by professionals in this field according to the actual situation and will not be analyzed here.

[0051] In the mode optimization module, if the predicted output value is less than the preset output threshold, no interaction mode will be performed. If the predicted output value is greater than the preset output threshold, the interaction mode will be improved.

[0052] The greater the predicted output value, the greater the light intensity proportionality coefficient, the motion speed proportionality coefficient, and the noise intensity proportionality coefficient, and the higher the interaction response ability; the smaller the predicted output value, the smaller the light intensity proportionality coefficient, the motion speed proportionality coefficient, and the noise intensity proportionality coefficient, and the lower the interaction response ability.

[0053] The method to improve the interaction mode is to increase the speech recognition accuracy to reach the preset speech recognition accuracy interval, and the speech recognition accuracy is improved through the adaptive filtering algorithm in the policy decision module.

[0054] The above formulas are all dimensionless and take their numerical calculations. The formula is obtained by collecting a large amount of data for software simulation to get a formula closest to the real situation. The preset parameters in the formula are set by technicians in this field according to the actual situation.

[0055] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions described in the embodiments of the present application are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center in a wired or wireless (such as infrared, wireless, microwave, etc.) manner. The computer-readable storage medium can be any available medium that can be accessed by a computer, or a data storage device such as a server or data center that includes one or more collections of available media. The available media can be magnetic media (such as floppy disks, hard disks, magnetic tapes), optical media (such as DVDs), or semiconductor media. The semiconductor media can be a solid-state drive.

[0056] It should be understood that in various embodiments of the present application, the order numbers of the above processes do not indicate the order of execution. The order of execution of each process should be determined by its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present application.

[0057] Those of ordinary skill in the art can realize that the units and algorithm steps of the examples described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present application.

[0058] Those skilled in the art can clearly understand that for the convenience and conciseness of description, the specific working processes of the systems, devices, and units described above can refer to the corresponding processes in the foregoing method embodiments, and will not be described herein again.

[0059] In several embodiments provided in this application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections to each other can be through some interfaces. The indirect couplings or communication connections of devices or units can be in electrical, mechanical, or other forms.

[0060] The units described as separate components may or may not be physically separated. The components displayed as units may or may not be physical units, that is, they can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0061] In addition, in each embodiment of this application, the functional units can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit.

[0062] If the functions are implemented in the form of software function units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in each embodiment of this application. The aforementioned storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs that can store program codes.

[0063] The above is only the specific implementation manner of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art within the technical scope disclosed in this application can easily think of changes or substitutions, which should all be covered within the protection scope of this application.

Claims

1. An intelligent interaction system applied to AI glasses, characterized in that: It includes a data acquisition module, an intention recognition module, a strategy decision-making module, and a mode optimization module. The functions of each module are as follows: The data acquisition module collects head pose data through an inertial measurement unit integrated inside the AI glasses, and calculates the head pose change amount based on the head pose data; The intention recognition module analyzes the head pose change amount and judges the complexity of the interaction scenario; The strategy decision-making module determines whether it is necessary to adjust the interaction strategy according to the complexity of the interaction scenario. If the complexity of the interaction scenario is low, system resource optimization is performed; if the complexity of the interaction scenario is high, the interaction response ability is enhanced, and the resource optimization result is evaluated based on the interaction response ability; The mode optimization module determines whether it is necessary to improve the interaction mode according to the system resource optimization result.

2. The intelligent interaction system applied to AI glasses according to claim 1, wherein: An X, Y, Z three-axis coordinate system is constructed with the geometric center of the inertial measurement unit integrated inside the AI glasses as the origin. The head pose data is the linear acceleration and angular velocity along the X, Y, Z three-axis coordinate system measured by the inertial measurement unit; Taking the current moment as the starting point, a fixed time period is selected forward as the acquisition period, the angular velocities of the X, Y, Z three-axis coordinate systems are collected, and the modulus length of the angular velocities is calculated to obtain the rotation synthesis value; Select the same fixed time period before the acquisition period as the previous acquisition period, collect the linear accelerations in the previous sampling period and the acquisition period respectively and take the difference, and then calculate the modulus length of the difference to obtain the acceleration change intensity.

3. The intelligent interaction system applied to AI glasses according to claim 2, wherein: Select a fixed number of the same fixed time periods before the acquisition period; Obtain the maximum rotation synthesis value and acceleration change intensity within all the same fixed time periods; Divide the rotation synthesis value and acceleration change intensity within the acquisition period by the maximum rotation synthesis value and acceleration change intensity within all the same fixed time periods respectively to obtain the normalized values; Fuse the normalized values according to the preset weighting coefficients to obtain the head pose change amount.

4. The intelligent interaction system applied to AI glasses according to claim 3, wherein: Obtain the position coordinates of the head, and expand to a spherical area with a fixed diameter around the position coordinates of the head in all directions in space as the sample range; Obtain all the head pose change amounts within the sample range, and merge the head pose change amounts into a change amount data set; Arrange the head pose change amount data in the change amount data set in ascending order, and obtain the median of the head pose change amounts in the change amount data set; Calculate the absolute deviation of each head pose change amount from the median to form an absolute deviation set; Take the median of the absolute deviation set to obtain the MAD value.

5. The intelligent interaction system applied to AI glasses according to claim 4, wherein: Compare the MAD value with the judgment threshold of the complexity of the interaction scenario; When the MAD value is greater than the judgment threshold of the complexity of the interaction scenario, it is determined that the complexity of the interaction scenario is high; When the MAD value is less than or equal to the judgment threshold of the complexity of the interaction scenario, it is determined that the complexity of the interaction scenario is low.

6. The intelligent interaction system applied to AI glasses according to claim 5, wherein: If the complexity of the interaction scenario is low, directly perform system resource optimization; If the complexity of the interaction scenario is high, improve the interaction response ability and then perform system resource optimization; When the complexity of the interaction scenario is low, obtain the angular velocity and acceleration of the head posture, and combine the operating frequency of the system to adjust the response until the head posture change amount reaches the head posture change amount threshold to achieve resource optimization.

7. The intelligent interaction system applied to AI glasses according to claim 5, wherein: When the complexity of the interaction scenario is high, obtain the speech recognition accuracy, noise intensity, and ambient light intensity within the sample range to comprehensively judge the interaction response ability; If the predicted speech recognition accuracy is not within the preset speech recognition accuracy interval, improve the speech recognition accuracy through an adaptive filtering algorithm; Combine the ambient light intensity into a light intensity data set, and take the median of the ambient light intensity as the light intensity threshold; Successively divide the ambient light intensity by the light intensity threshold to obtain each light intensity coefficient.

8. The intelligent interaction system applied to AI glasses according to claim 7, wherein: When the light intensity coefficient is within the preset light intensity coefficient interval threshold, retain the corresponding ambient light intensity and take the average to obtain the average light intensity coefficient; Take the ratio of the light intensity of the predicted interaction scenario to the average light intensity coefficient as the light intensity ratio coefficient; Obtain the motion speed data within the sample range, the angular velocity of the predicted interaction scenario, and the angular velocity value of the sample range, and comprehensively calculate to obtain the motion speed ratio coefficient; Take the ratio of the noise intensity to the preset noise intensity threshold as the noise intensity ratio coefficient.

9. The intelligent interaction system applied to AI glasses according to claim 8, wherein: Construct a long short-term memory network model through the light intensity ratio coefficient, motion speed ratio coefficient, and noise intensity ratio coefficient to obtain the predicted output value; If the predicted output value is less than the preset output threshold, do not improve the interaction mode; If the predicted output value is greater than the preset output threshold, improve the interaction mode.

Images