Eye movement interaction intention recognition dynamic optimization method for shaking environment

Abstract

The invention discloses an eye movement interaction intention recognition dynamic optimization method for a shaking environment, and belongs to the field of human-computer interaction and human factor engineering. According to the method, the eye movement tracking and machine learning technologies are adopted, the interaction intention recognition optimization model is constructed, the training efficiency of the interaction intention recognition optimization model is improved, a user does not need to deliberately change the natural sight movement habit when using eye movement interaction, the naturalness and intuitiveness of interaction are improved, and the user experience is improved. Therefore, the user can more quickly adapt to and master eye movement interaction, the learning cost is reduced, and the satisfaction degree and the comfort degree of the user when the user uses the eye movement interaction are enhanced. According to the method, shaking environments are parameterized and aggregated into shaking factors, a mapping relation between the shaking factors and an optimal classification threshold value of the eye movement interaction intention recognition model is established, dynamic adjustment of the accuracy of the eye movement interaction intention recognition model in different shaking environments is achieved, and the method can adapt to various different shaking conditions; the negative influence caused by the shaking environment is reduced or compensated, and the accuracy and reliability of eye movement interaction in the shaking environment are ensured.

Description

Technical Field

[0001] The invention relates to a dynamic optimization method for eye movement interaction intention recognition in a shaking environment, and belongs to the field of human-computer interaction and ergonomics. Background Art

[0002] As an emerging interaction paradigm, eye movement interaction technology has been increasingly valued for its naturalness and efficiency. The core of this technology is to imitate the most instinctive way of communication of human beings - visual gaze, allowing users to communicate information with devices without physical contact. This way of communication not only reduces the need for contact in traditional interaction modes, but also reduces the potential health risks caused by contact, while improving the convenience and concealment of interaction. Although eye movement interaction technology has shown great potential, its implementation still faces challenges. Traditional eye movement interaction systems mostly adopt rule-based explicit interaction design, which requires users to perform specific and unnatural eye movements to issue instructions, which is significantly different from users' expectations for natural interaction experience. As the human-centered design concept gradually dominates the field of interaction design, users have put forward higher requirements for the naturalness and intuitiveness of the interaction experience. Users expect eye movement interaction technology to provide an interactive experience that does not require additional learning costs and is consistent with daily communication habits. This experience should be able to seamlessly integrate into the user's natural behavior, allowing users to complete the interaction with the device unconsciously, thereby achieving truly natural interaction.

[0003] The development of machine learning algorithms has brought revolutionary progress to eye movement interaction technology. By collecting users' eye movement data in real time, the algorithm can mine key eye movement features that can accurately characterize users' interaction intentions. These features include but are not limited to the number of gazes, gaze duration, eye movement speed, and eye movement trajectory. Using these features, a complex eye movement interaction intention recognition model can be constructed, which can identify users' interaction intentions with high accuracy and efficiency, thereby providing a natural and efficient eye movement interaction experience. The adaptive learning ability of machine learning algorithms enables the system to continuously optimize over time, better understand users' personalized interaction habits, and further improve the naturalness and efficiency of interaction.

[0004] Although eye movement interaction technology has made significant progress in laboratory environments, the accuracy and stability of eye movement interaction intention recognition face great challenges in practical applications, especially in shaky environments. Shaky environments can cause a lot of noise and outliers in eye movement data, and these interference factors can seriously affect the accuracy of eye movement interaction intention recognition. For example, in scenarios such as vehicle driving, sea voyages, or air flights, the user's eye movement data will be subject to continuous external interference, which makes it difficult for traditional eye movement interaction models to adapt. Therefore, how to design an eye movement interaction intention recognition method that can maintain high accuracy in a complex and shaky real environment has become an important part of the development of eye movement interaction technology.

[0005] Therefore, applying new technologies to the field of human-computer interaction and ergonomics and dynamically optimizing eye movement interaction intention recognition methods based on shaking environments has become an urgent problem to be solved. Summary of the invention

[0006] The purpose of the present invention is to provide a dynamic optimization method for eye movement interaction intention recognition in a shaking environment, dynamically optimize the accuracy of the intention recognition model in different shaking environments, and achieve an efficient eye movement interaction experience.

[0007] The objective of the present invention is achieved through the following technical solutions:

[0008] The present invention discloses a dynamic optimization method for eye movement interaction intention recognition in a shaking environment, comprising the following steps:

[0009] Step 1: Preset multiple groups of parameter combinations consisting of six degrees of freedom (lateral displacement, longitudinal displacement, vertical displacement, roll angle, offset angle, pitch angle) and motion cycle parameters, as well as conversion cycles; simulate corresponding shaking environments according to different parameter combinations. The motion cycle is the setting time of a certain set of parameters, and the conversion cycle is the switching time of different parameter combinations. In different shaking environments (including static environments), eye movement interaction selection experiments are conducted to collect raw eye movement data and subject interaction intention labels. The raw eye movement data includes the x and y position coordinates of the left and right eye gaze points in the two-dimensional screen, the left and right eye pupil diameters, and the left and right eyelid openings. The subject interaction intention labels include 0 and 1, 0 means no intention, and 1 means intention. The eye movement behavior detection algorithm is used to distinguish between fixation and eye saccade behavior, and the eye movement features are calculated. The eye movement features and the subject interaction intention labels together constitute the eye movement interaction intention recognition data set. Eye movement features include fixation time, number of fixations, and number of eye saccades.

[0010] Step 2: Use the eye movement interaction intention recognition dataset obtained in a static environment to build a basic model for eye movement interaction intention recognition, and establish a mapping relationship between eye movement features and interaction intention. Interaction intention is divided into two categories: no intention and intention, which correspond to the interaction intention labels 0 and 1 in step 1 respectively. Using the eye movement interaction intention recognition dataset obtained in step 1 as input, train the candidate models of the basic model for intention recognition respectively, and select the model with the highest accuracy from the trained candidate models as the basic model for eye movement interaction intention recognition.

[0011] Preferably, the candidate models of the basic model for intention recognition include SVM model, RF model, XGBoost model, and LR model. The eye movement interaction intention recognition dataset obtained in step one is used as input, and the SVM model, RF model, XGBoost model, and LR model are trained respectively. The model with the highest accuracy is selected from the trained SVM model, RF model, XGBoost model, and LR model as the basic model for eye movement interaction intention recognition.

[0012] Step 3: Aggregate the shaking parameters in step 1 into a shaking factor to reflect the severity of the shaking. The normalized velocities in the three translation directions are calculated. Similarly, The normalized angular velocity in the three rotation directions is calculated, where x represents the translation parameter, θ represents the rotation parameter, t represents the motion period, j∈{1,2,3} represents the lateral displacement, longitudinal displacement, and vertical displacement, respectively, k∈{1,2,3} represents the roll angle, offset angle, and pitch angle, respectively, i∈{1,2,3,...} represents the combined parameter conversion period sequence, and f norm Represents the normalization function. Using the formula v i =||v i ||2 Take the second norm of vector v to get the composite speed of the three translational velocities. Similarly, use the formula w i =||w i ||2 Take the second norm of vector w to get the composite angular velocity of the three rotational velocities, where Using the formula Δv i = | v i+1 -v i | ,Δw i = | w i+1 -w i | ,Δt i =t i+1 -t i | Calculate the velocity change, angular velocity change and conversion period change between two sets of adjacent parameters. Using the formula λ1 = fmean (Δv i )×(1+f mean (Δt i )) Calculate the sway factor component λ1, using the formula λ2=f mean (Δw i )×(1+f mean (Δt i )) Calculate the sway factor component λ2, where f mean is the average function. Finally, using the formula The sway factor λ is calculated, where λ=[λ1,λ2] T .

[0013] Step 4: Set the default value of the classification threshold of the eye movement interaction intention recognition basic model constructed in step 2, which is used for the intent category recognized by the eye movement interaction intention recognition basic model output. Take the candidate sequence of classification thresholds at equal intervals between 0 and 1, and change the classification threshold of the eye movement interaction intention recognition basic model constructed in step 2 according to this candidate sequence. Analyze the changing trend of the accuracy of the eye movement interaction intention recognition basic model, and determine the range of the optimal classification threshold of the eye movement interaction intention recognition basic model.

[0014] Preferably, the default value of the classification threshold of the eye movement interaction intention recognition basic model constructed in step 2 is set to 0.5.

[0015] Step 5: Use different shake factors and the classification threshold candidate sequence of the eye movement interaction intention recognition basic model constructed in step 2 as features, and the accuracy of the eye movement interaction intention recognition basic model constructed in step 2 as a label to establish an accuracy prediction dataset, train a polynomial regression model, and build an accuracy prediction model based on the trained polynomial regression model.

[0016] Step 6: Obtain different sway environments, basic model classification thresholds and corresponding basic model accuracy through the accuracy prediction model. The corresponding optimal basic model classification threshold is obtained by screening the optimal basic model accuracy, and the mapping relationship between the sway factor and the optimal classification threshold of the basic model is constructed. The method of constructing the mapping relationship between the sway factor and the optimal classification threshold of the basic model is called the optimal classification threshold prediction algorithm.

[0017] Step 7: According to the optimal classification threshold prediction algorithm in step 6, the optimal classification threshold of the eye movement interaction intention recognition basic model corresponding to the specific shaking environment is obtained, and the classification threshold of the eye movement interaction intention recognition basic model is adjusted according to the threshold and used as the eye movement interaction intention recognition optimization model. The eye movement features are used as the input of the eye movement interaction intention recognition optimization model to obtain the eye movement interaction intention recognition result, realize the dynamic optimization of eye movement interaction intention recognition in different shaking environments, reduce the negative impact of shaking on eye movement interaction intention recognition according to the eye movement interaction intention recognition result, and realize an efficient eye movement interaction experience.

[0018] Beneficial effects:

[0019] 1. The present invention discloses a dynamic optimization method for eye movement interaction intention recognition in a shaking environment, which adopts eye movement tracking and machine learning technology to construct an eye movement interaction intention recognition model, thereby reducing the preliminary training and learning required by users when using eye movement interaction technology. Users do not need to deliberately change their natural sight movement habits when using eye movement interaction, thereby improving the naturalness and intuitiveness of the interaction, allowing users to adapt to and master eye movement interaction more quickly, reducing learning costs, and enhancing user satisfaction and comfort when using eye movement interaction.

[0020] 2. The present invention discloses a dynamic optimization method for eye movement interaction intention recognition in a shaking environment, which parameterizes the shaking environment and aggregates it into shaking factors, establishes a mapping relationship between the shaking factors and the optimal classification threshold of the eye movement interaction intention recognition model, and realizes dynamic adjustment of the accuracy of the eye movement interaction intention recognition model under different shaking environments. It can adapt to various shaking conditions to reduce or compensate for the negative impact of the shaking environment, ensure the accuracy and reliability of eye movement interaction in a shaking environment, and broaden its application prospects in the fields of medicine, transportation, virtual reality, etc., providing new interaction solutions for the above-mentioned fields.

[0021] 3. The present invention discloses a dynamic optimization method for eye movement interaction intention recognition in a shaking environment, which presets multiple groups of parameter combinations consisting of six degrees of freedom of lateral displacement, longitudinal displacement, vertical displacement, roll angle, offset angle, pitch angle, motion cycle parameters, and conversion cycle; and simulates the corresponding shaking environment according to different parameter combinations. The motion cycle is the setting time of a certain group of parameters, and the conversion cycle is the switching time of different parameter combinations. In different shaking environments, an eye movement interaction selection experiment is carried out to collect raw eye movement data and subject interaction intention labels. The raw eye movement data includes the x and y position coordinates of the left and right eye gaze points in the two-dimensional screen, the left and right eye pupil diameters, and the left and right eyelid openings. The subject interaction intention labels include 0 and 1, 0 indicates no intention, and 1 indicates intention. The eye movement behavior detection algorithm is used to distinguish between fixation and eye saccade behavior, and the eye movement features are calculated. The eye movement features and the subject interaction intention labels together constitute the eye movement interaction intention recognition data set. The eye movement features include fixation time, fixation times, and eye saccade times. The eye movement features and the subject interaction intention labels together constitute the eye movement interaction intention recognition data set. The eye movement interaction intention recognition dataset constructed by the present invention can improve the accuracy and efficiency of the eye movement interaction intention recognition optimization model.

[0022] 4. The present invention discloses a dynamic optimization method for eye movement interaction intention recognition in a shaking environment. The candidate models of the basic model for intention recognition include SVM model, RF model, XGBoost model, and LR model. The SVM model, RF model, XGBoost model, and LR model are trained separately. The model with the highest accuracy is selected from the trained SVM model, RF model, XGBoost model, and LR model as the basic model for eye movement interaction intention recognition, which can improve the accuracy and efficiency of dynamic optimization of eye movement interaction intention recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 A schematic diagram of constructing an accuracy prediction model for a dynamic optimization method of eye movement interaction intention recognition in a shaking environment according to an embodiment of the present invention.

[0024] Figure 2 Schematic diagram of an optimal classification threshold prediction algorithm for a dynamic optimization method for eye movement interaction intention recognition in a shaking environment according to an embodiment of the present invention. DETAILED DESCRIPTION

[0025] In order to better illustrate the purpose and advantages of the present invention, the invention is further described below with reference to the accompanying drawings and examples.

[0026] Embodiment 1:

[0027] The embodiment applies a dynamic optimization method for eye movement interaction intention recognition in a shaking environment of the present invention to dynamically optimize eye movement interactions in two different shaking environments.

[0028] This embodiment discloses a dynamic optimization method for eye movement interaction intention recognition in a shaking environment, and the specific implementation steps are as follows:

[0029] Step 1: Preset four groups of parameter combinations consisting of six degrees of freedom (lateral displacement, longitudinal displacement, vertical displacement, roll angle, offset angle, pitch angle) and motion cycle parameters, as well as conversion cycle; simulate the corresponding shaking environment according to different parameter combinations, and the shaking parameters are shown in Table 1. The motion cycle is the setting time of a certain group of parameters, and the conversion cycle is the switching time of different parameter combinations. Conduct eye movement interaction selection experiments in three different environments: static, low shaking, and high shaking, and collect raw eye movement data and subject interaction intention labels. The raw eye movement data includes the x and y position coordinates of the left and right eye gaze points in the two-dimensional screen, the left and right eye pupil diameters, and the left and right eyelid openings. The subject interaction intention labels include 0 and 1, 0 means no intention, and 1 means intention. The I-VT eye movement behavior detection algorithm is used to distinguish between fixation and eye saccade behavior, and the threshold parameters are shown in Table 2. The eye movement features are calculated, and the eye movement features and the subject interaction intention labels together constitute the eye movement interaction intention recognition data set. The eye movement features include fixation time, fixation times, and eye saccade times, as shown in Table 3.

[0030] Table 1 Sway parameter table

[0031]

[0032] Step 2: Use the eye movement interaction intention recognition dataset obtained in a static environment to build a basic model for eye movement interaction intention recognition, and establish a mapping relationship between eye movement features and interaction intention. Interaction intention is divided into two categories: no intention and intention, which correspond to the interaction intention labels 0 and 1 in step 1 respectively. Using the eye movement interaction intention recognition dataset obtained in step 1 as input, train the candidate models of the basic model for intention recognition respectively, and select the model with the highest accuracy from the trained candidate models as the basic model for eye movement interaction intention recognition.

[0033] Preferably, the candidate models of the basic model for intention recognition include SVM model, RF model, XGBoost model, and LR model. The eye movement interaction intention recognition dataset obtained in step one is used as input, and the SVM model, RF model, XGBoost model, and LR model are trained respectively. The model with the highest accuracy is selected from the trained SVM model, RF model, XGBoost model, and LR model as the basic model for eye movement interaction intention recognition.

[0034] Table 2 I-VT algorithm threshold parameter table

[0035]

[0036] Table 3 Eye movement characteristics

[0037]

[0038] Step 3: Aggregate the shaking parameters in step 1 into a shaking factor to reflect the severity of the shaking. The normalized velocities in the three translation directions are calculated. Similarly, The normalized angular velocity in the three rotation directions is calculated, where x represents the translation parameter, θ represents the rotation parameter, t represents the motion period, j∈{1,2,3} represents the lateral displacement, longitudinal displacement, and vertical displacement, respectively, k∈{1,2,3} represents the roll angle, offset angle, and pitch angle, respectively, i∈{1,2,3,...} represents the combined parameter conversion period sequence, and f norm Represents the normalization function. Using the formula v i =||v i ||2 Take the second norm of vector v to get the composite speed of the three translational velocities. Similarly, use the formula w i =||w i ||2 Take the second norm of vector w to get the composite angular velocity of the three rotational velocities, where Using the formula Δv i =|v i+1 -v i |,Δw i =|w i+1 -w i |,Δt i =|t i+1 -t i | Calculate the velocity change, angular velocity change, and conversion period change between two sets of adjacent parameters. Using the formula λ1=f mean (Δv i )×(1+f mean (Δt i )) Calculate the sway factor component λ1, using the formula λ2=f mean (Δw i )×(1+f mean (Δt i )) Calculate the sway factor component λ2, where f mean is the average function. Finally, using the formula The sway factor λ is calculated, where λ=[λ1,λ2] T .

[0039] Step 4: Set the default value of the classification threshold of the eye movement interaction intention recognition basic model constructed in step 2, which is used for the intent category recognized by the eye movement interaction intention recognition basic model output. Take the candidate sequence of classification thresholds at equal intervals between 0 and 1 with a step length of 0.005, and change the classification threshold of the eye movement interaction intention recognition basic model constructed in step 2 according to this candidate sequence. Analyze the changing trend of the accuracy of the eye movement interaction intention recognition basic model, and determine that the optimal classification threshold of the eye movement interaction intention recognition basic model is in the range of 0.5 to 0.6.

[0040] Preferably, the default value of the classification threshold of the eye movement interaction intention recognition basic model constructed in step 2 is set to 0.5.

[0041] Step 5: If Figure 1 As shown, different shake factors and the classification threshold candidate sequence of the basic model for eye movement interaction intention recognition constructed in step 2 are used as features, and the accuracy of the basic model for eye movement interaction intention recognition constructed in step 2 is used as a label to establish an accuracy prediction data set, and a polynomial regression model is trained, and an accuracy prediction model is constructed based on the trained polynomial regression model.

[0042] Step 6: Obtain different sway environments, basic model classification thresholds and corresponding basic model accuracy through the accuracy prediction model. The corresponding optimal basic model classification threshold is obtained by screening the optimal basic model accuracy, and the mapping relationship between the sway factor and the optimal classification threshold of the basic model is constructed. The method of constructing the mapping relationship between the sway factor and the optimal classification threshold of the basic model is called the optimal classification threshold prediction algorithm.

[0043] Step 7: If Figure 2 As shown, according to the optimal classification threshold prediction algorithm in step six, the optimal classification threshold of the eye movement interaction intention recognition basic model corresponding to the specific shaking environment is obtained, and the classification threshold of the eye movement interaction intention recognition basic model is adjusted according to the threshold and used as the eye movement interaction intention recognition optimization model. The eye movement features are used as the input of the eye movement interaction intention recognition optimization model to obtain the eye movement interaction intention recognition result, realize the dynamic optimization of eye movement interaction intention recognition in different shaking environments, reduce the negative impact of shaking on eye movement interaction intention recognition according to the eye movement interaction intention recognition result, and realize an efficient eye movement interaction experience.

[0044] The specific description above further illustrates the purpose, technical solutions and beneficial effects of the invention in detail. It should be understood that the above is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A dynamic optimization method for eye movement interaction intention recognition in a shaking environment, characterized by: The following steps are included: Step 1: Preset multiple groups of parameter combinations consisting of six degrees of freedom, including lateral displacement, longitudinal displacement, vertical displacement, roll angle, offset angle, pitch angle, motion cycle parameters, and conversion cycle; simulate corresponding shaking environments according to different parameter combinations; the motion cycle is the setting time of a certain group of parameters, and the conversion cycle is the switching time of different parameter combinations; conduct eye movement interaction selection experiments in different shaking environments, and collect raw eye movement data and subject interaction intention labels; the raw eye movement data includes the x and y position coordinates of the left and right eye gaze points in the two-dimensional screen, the left and right eye pupil diameters, and the left and right eyelid openings, and the subject interaction intention labels include 0 and 1, 0 means no intention, and 1 means intention; distinguish between fixation and eye saccade behavior through an eye movement behavior detection algorithm, calculate eye movement features, and the eye movement features and subject interaction intention labels together constitute an eye movement interaction intention recognition data set; Step 2: Use the eye movement interaction intention recognition dataset obtained in a static environment to build a basic model for eye movement interaction intention recognition, and establish a mapping relationship between eye movement features and interaction intentions; The interaction intention is divided into two categories: no intention and intention, which correspond to the interaction intention labels 0 and 1 in step one respectively; the candidate models of the basic model for intention recognition include SVM model, RF model, XGBoost model, and LR model. The eye movement interaction intention recognition dataset obtained in step one is used as input to train the SVM model, RF model, XGBoost model, and LR model respectively, and the model with the highest accuracy is selected from the trained SVM model, RF model, XGBoost model, and LR model as the basic model for eye movement interaction intention recognition; Step 3: Aggregate the shaking parameters in step 1 into a shaking factor to reflect the severity of the shaking; The normalized velocities in the three translation directions are calculated by The normalized angular velocity in the three rotation directions is calculated, where x represents the translation parameter, θ represents the rotation parameter, t represents the motion period, j∈{1,2,3} represents the lateral displacement, longitudinal displacement, and vertical displacement, respectively, k∈{1,2,3} represents the roll angle, offset angle, and pitch angle, respectively, i∈{1,2,3,...} represents the combined parameter conversion period sequence, and f norm Represents the normalization function; using the formula v i =||v i ||2 Take the second norm of vector v to get the composite speed of the three translational velocities, and use the formula w i =||w i ||2 Take the second norm of vector w to get the composite angular velocity of the three rotational velocities, where Using the formula Δv i =|v i+1 -v i |,Δw i =|w i+1 -w i |,Δt i =|t i+1 -t i |Calculate the velocity change, angular velocity change and conversion period change between two sets of adjacent parameters; Using the formula λ1=f mean (Δv i )×(1+f mean (Δt i )) Calculate the sway factor component λ1, using the formula λ2=f mean (Δw i )×(1+f mean (Δt i )) Calculate the sway factor component λ2, where f mean is the average function; using the formula The sway factor λ is calculated, where λ=[λ1,λ2] T ; Step 4: Set the default value of the classification threshold of the eye movement interaction intention recognition basic model constructed in step 2, which is used for the intent category recognized by the output of the eye movement interaction intention recognition basic model; obtain a candidate sequence of classification thresholds at equal intervals between 0 and 1, and change the classification threshold of the eye movement interaction intention recognition basic model constructed in step 2 according to the candidate sequence; analyze the changing trend of the accuracy of the eye movement interaction intention recognition basic model, and determine the range of the optimal classification threshold of the eye movement interaction intention recognition basic model; Step 5: Use different shaking factors and the classification threshold candidate sequence of the eye movement interaction intention recognition basic model constructed in step 2 as features, and the accuracy of the eye movement interaction intention recognition basic model constructed in step 2 as a label to establish an accuracy prediction data set, train a polynomial regression model, and construct an accuracy prediction model based on the trained polynomial regression model; Step 6: Obtain different sway environments, basic model classification thresholds and corresponding basic model accuracy through the accuracy prediction model; obtain the corresponding optimal basic model classification threshold by screening the optimal basic model accuracy, and construct a mapping relationship between the sway factor and the optimal classification threshold of the basic model. The method of constructing the mapping relationship between the sway factor and the optimal classification threshold of the basic model is called the optimal classification threshold prediction algorithm; Step seven: According to the optimal classification threshold prediction algorithm in step six, the optimal classification threshold of the eye movement interaction intention recognition basic model corresponding to the specific shaking environment is obtained, and the classification threshold of the eye movement interaction intention recognition basic model is adjusted according to the threshold and used as the eye movement interaction intention recognition optimization model; the eye movement features are used as the input of the eye movement interaction intention recognition optimization model to obtain the eye movement interaction intention recognition results, and the eye movement interaction intention recognition is dynamically optimized in different shaking environments. According to the eye movement interaction intention recognition results, the negative impact of shaking on the eye movement interaction intention recognition is reduced to achieve an efficient eye movement interaction experience.

2. The method for dynamic optimization of eye movement interaction intention recognition in a shaking environment as claimed in claim 1, characterized in that: Eye movement characteristics include fixation time, number of fixations, and number of saccades.

3. A method for dynamic optimization of eye movement interaction intention recognition in a shaking environment as claimed in claim 1 or 2, characterized in that: In step 2, the candidate models of the basic model for intention recognition include SVM model, RF model, XGBoost model, and LR model. The eye movement interaction intention recognition dataset obtained in step 1 is used as input, and the SVM model, RF model, XGBoost model, and LR model are trained respectively. The model with the highest accuracy is selected from the trained SVM model, RF model, XGBoost model, and LR model as the basic model for eye movement interaction intention recognition.

4. The method for dynamic optimization of eye movement interaction intention recognition in a shaking environment as claimed in claim 3, characterized in that: In step three, the default value of the classification threshold of the eye movement interaction intention recognition basic model constructed in step two is set to 0.5.

Images