Driving ability analysis system for stroke patients based on eye tracking technology

Abstract

The present application discloses a driving ability analysis system for stroke patients based on eye tracking technology, which relates to the technical field of driving ability analysis for stroke patients. It can collect eye state tracking videos through eye tracking equipment, and introduce data processing and image analysis algorithms based on artificial intelligence and machine vision analysis at the back end to analyze the eye state tracking videos, so as to learn and characterize the temporal propagation aggregation representation characteristics of the eye behavior pattern of stroke patients, so as to reflect the semantic information such as the eye gaze point position, gaze time, pupil size change of stroke patients, so as to evaluate and detect whether the patient is suitable for driving. In this way, by combining eye tracking technology with advanced artificial intelligence and deep learning algorithms, it is possible to realize intelligent analysis and objective evaluation of the driving ability of stroke patients, so as to judge whether the driver is suitable for driving, and provide scientific basis for medical professionals.

Description

Technical Field

[0001] The present application relates to the technical field of driving ability analysis of stroke patients, and more specifically, to a driving ability analysis system for stroke patients based on eye tracking technology. Background Art

[0002] Stroke (commonly known as stroke) is a common neurological disease that can seriously affect patients' cognitive and physical functions. For many stroke survivors, regaining the ability to drive independently is an important rehabilitation goal because it is directly related to whether they can regain autonomy in life. However, after experiencing brain damage, stroke may cause problems such as distraction and visual spatial perception disorders, which may make it impossible to identify traffic signs, pedestrians or other vehicles in time during driving, directly affecting driving safety. Therefore, it is necessary to analyze and judge whether stroke patients have the ability to drive safely.

[0003] However, the traditional analysis of driving ability of stroke patients mainly relies on clinical evaluation and standardized tests. Although these methods can reflect the patient's physical condition and cognitive function to a certain extent, they also have some obvious defects. For example, the traditional analysis of driving ability of stroke patients often needs to rely on the experience and personal judgment of doctors or evaluators, which may lead to the analysis results of driving ability of stroke patients being highly subjective and lacking unified standards. At the same time, the traditional driving ability analysis and evaluation process often requires multiple face-to-face clinical consultations and long-term observations, which not only consumes a lot of time and human resources, but also increases the economic burden on patients.

[0004] Therefore, an optimized driving ability analysis system for stroke patients is desired. Summary of the invention

[0005] The present application provides a driving ability analysis system for stroke patients based on eye tracking technology. By combining eye tracking technology with advanced artificial intelligence and deep learning algorithms, it can realize intelligent analysis and objective evaluation of the driving ability of stroke patients, so as to judge whether the driver is suitable for driving and provide a scientific basis for medical professionals.

[0006] In the first aspect, a driving ability analysis system for stroke patients based on eye tracking technology is provided, comprising:

[0007] An eyeball state tracking video acquisition module is used to acquire the eyeball state tracking video acquired by the eyeball tracking device;

[0008] A key frame sampling module, used for performing key frame sampling on the eye state tracking video to obtain a time queue of eye state tracking key frames;

[0009] An eye state feature extraction module, used for extracting eye state features from each eye state tracking key frame in the time queue of the eye state tracking key frames to obtain a time queue of eye state semantic feature vectors;

[0010] A feature filtering processing module, used for performing feature filtering processing based on contribution semantics on the time queue of the eyeball state semantic feature vector to obtain a filtered time queue of the eyeball state semantic feature vector;

[0011] Wherein, the feature filtering processing module includes: a cluster analysis unit, which is used to perform cluster analysis on the time queue of the eye state semantic feature vector to obtain an eye state semantic cluster center representation vector; a semantic contribution calculation unit, which is used to calculate the semantic contribution of each eye state semantic feature vector in the time queue of the eye state semantic feature vector relative to the eye state semantic cluster center representation vector to obtain a sequence of masked sparse eye state semantic contribution representation matrices; a feature filtering calculation unit, which is used to use each masked sparse eye state semantic contribution representation matrix in the sequence of masked sparse eye state semantic contribution representation matrices as a modulation matrix, and respectively calculate the matrix product between it and each eye state semantic feature vector to obtain the time queue of the filtered eye state semantic feature vector;

[0012] A feature message propagation module, used for performing feature message propagation on the time queue of the filtered eye state semantic feature vector to obtain an eye behavior pattern time series propagation aggregation representation vector;

[0013] The driving ability analysis module is used to perform driving ability analysis based on the eye behavior pattern temporal propagation aggregation representation vector to determine whether the driver is suitable for driving.

[0014] In a possible implementation, the eye state feature extraction module is used to: input each eye state tracking key frame in the time queue of the eye state tracking key frame into an eye state feature extractor based on a void convolutional neural network model to obtain the time queue of the eye state semantic feature vector.

[0015] In a possible implementation, the cluster analysis unit is used to calculate the positional mean between each eye state semantic feature vector in the time queue of the eye state semantic feature vector to obtain the eye state semantic cluster center representation vector.

[0016] In a possible implementation, the semantic contribution calculation unit includes:

[0017] A semantic contribution representation matrix calculation subunit is used to use the eye state semantic cluster center representation vector as a pseudo anchor center, and calculate the semantic contribution representation matrix of each eye state semantic feature vector in the time queue of the eye state semantic feature vector relative to the pseudo anchor center to obtain a sequence of eye state semantic contribution representation matrices;

[0018] The random deactivation processing subunit is used to perform random deactivation processing on each eye state semantic contribution representation matrix in the sequence of eye state semantic contribution representation matrices to obtain the sequence of masked sparse eye state semantic contribution representation matrices.

[0019] In a possible implementation, the semantic contribution representation matrix calculation subunit is used to:

[0020] Calculating the positional division between the transposed vector of the eye state semantic cluster center representation vector and the bi-norm of the eye state semantic cluster center representation vector to obtain a transformed eye state semantic cluster center vector;

[0021] After calculating the vector multiplication between each eye state semantic feature vector and the transformed eye state semantic cluster center vector, use The function performs soft maximum normalization processing on the obtained sequence of association matrices to obtain the sequence of eye state semantic contribution representation matrices.

[0022] In a possible implementation, the feature message propagation module includes:

[0023] a current filtered feature vector extraction unit, configured to extract a current filtered eye state semantic feature vector from the time queue of the filtered eye state semantic feature vectors, and define other filtered eye state semantic feature vectors in the time queue of the filtered eye state semantic feature vectors as historical filtered eye state semantic feature vectors to obtain a time queue of historical filtered eye state semantic feature vectors;

[0024] An eye state semantic static energy value calculation unit, used for calculating the eye state semantic static energy value of each historical filtered eye state semantic feature vector in the time queue of the historical filtered eye state semantic feature vector;

[0025] An eye state semantic energy spatial attenuation factor determination unit, used to determine the eye state semantic energy spatial attenuation factor based on the spatial span coefficient between the eye state semantic feature vectors after each historical filtering and the eye state semantic feature vector after the current filtering;

[0026] An eye state semantic energy time decay factor determination unit, used to determine the eye state semantic energy time decay factor based on the time span coefficient between the eye state semantic feature vectors after each historical filtering and the eye state semantic feature vector after the current filtering;

[0027] An eye state semantic dynamic energy spatiotemporal transfer value calculation unit is used to perform spatiotemporal modulation on the eye state semantic static energy value of each of the historically filtered eye state semantic feature vectors based on the eye state semantic energy spatial attenuation factor and the eye state semantic energy temporal attenuation factor to obtain the eye state semantic dynamic energy spatiotemporal transfer value of each of the historically filtered eye state semantic feature vectors relative to the currently filtered eye state semantic feature vector;

[0028] An eye state semantic dynamic energy spatiotemporal transfer weight calculation unit is used to input the eye state semantic dynamic energy spatiotemporal transfer value of each of the historically filtered eye state semantic feature vectors into an energy transfer gating module based on a mask function to obtain a sequence of eye state semantic dynamic energy spatiotemporal transfer weights;

[0029] A transfer aggregation modulation vector calculation unit is used to calculate the weighted sum of the time queue of the eye state semantic feature vector after historical filtering based on the sequence of the eye state semantic dynamic energy spatiotemporal transfer weights to obtain the eye state semantic node feature transfer aggregation modulation vector after historical filtering;

[0030] A pattern time series propagation aggregate representation vector calculation unit, used for calculating the positional sum of the eye state semantic node feature transfer aggregate modulation vector after historical filtering and the eye state semantic feature vector after current filtering to obtain the eye behavior pattern time series propagation aggregate representation vector;

[0031] Wherein, the eye state semantic energy space attenuation factor determination unit is used to:

[0032] Calculating the spatial distance span between the eye state semantic feature vector after historical filtering and the eye state semantic feature vector after current filtering to obtain the eye state semantic spatial span coefficient after historical filtering;

[0033] The eye state semantic static energy value corresponding to the eye state semantic feature vector after historical filtering is divided by the eye state semantic space span coefficient after historical filtering to obtain the eye state semantic energy space attenuation factor.

[0034] In a possible implementation, the eye state semantic static energy value calculation unit is used to:

[0035] Calculating the mean and standard deviation of the eye state semantic feature vector after historical filtering to obtain the mean of the eye state semantic feature after historical filtering and the standard deviation of the eye state semantic feature after historical filtering;

[0036] After calculating the fourth power of the position difference between the historical filtered eye state semantic feature vector and the historical filtered eye state semantic feature mean, the fourth power is calculated to modulate the expected value of the historical filtered eye state semantic offset vector to obtain the filtered eye state semantic feature offset expected factor;

[0037] After calculating the division between the expected factor of the filtered eye state semantic feature offset and the fourth power of the historical filtered eye state semantic standard deviation, it is added to the maximum eigenvalue in the historical filtered eye state semantic feature vector to obtain the eye state semantic static energy value.

[0038] In a possible implementation, the eye state semantic energy time decay factor determination unit is used to:

[0039] After calculating the subtraction between the timestamp of the currently filtered eye state semantic feature vector and the timestamp of the historically filtered eye state semantic feature vector, the obtained timestamp value is rounded down to obtain a filtered eye state semantic time span coefficient;

[0040] After dividing the filtered eye state semantic time span coefficient by the inverse scaling parameter controlling the time decay period, the obtained time span decay representation coefficient is used as an exponential power, and a natural exponential function value with a natural constant e as the base is calculated to obtain a time span decay class support representation coefficient;

[0041] The division between the eye state semantic static energy value and the time span attenuation class support representation coefficient is calculated to obtain the eye state semantic energy time attenuation factor.

[0042] In a possible implementation, the eye state semantic dynamic energy spatiotemporal transfer weight calculation unit is used to: in the energy transfer gating module based on the mask function, in response to the eye state semantic dynamic energy spatiotemporal transfer value being greater than a threshold hyperparameter, The function normalizes the eye state semantic dynamic energy spatiotemporal transfer value that is greater than the threshold hyperparameter to obtain the eye state semantic dynamic energy spatiotemporal transfer weight.

[0043] In a possible implementation, the driving ability analysis module is used to: input the eye behavior pattern temporal propagation aggregation representation vector into a classifier-based driving ability analysis module to obtain an analysis result, and the analysis result is used to indicate whether the person is suitable for driving.

[0044] The present application provides a driving ability analysis system for stroke patients based on eye tracking technology, which can collect eye state tracking videos through eye tracking equipment, and introduce data processing and image analysis algorithms based on artificial intelligence and machine vision analysis at the back end to analyze the eye state tracking videos, so as to learn and characterize the temporal propagation aggregation representation characteristics of the eye behavior patterns of stroke patients, which are used to reflect the semantic information such as the eye gaze point position, gaze time, pupil size change of stroke patients, so as to evaluate and detect whether the patient is suitable for driving. In this way, by combining eye tracking technology with advanced artificial intelligence and deep learning algorithms, it is possible to realize intelligent analysis and objective evaluation of the driving ability of stroke patients, so as to judge whether the driver is suitable for driving and provide a scientific basis for medical professionals. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] The above and other purposes, features and advantages of the present invention will become more apparent by describing the embodiments of the present invention in more detail in conjunction with the accompanying drawings. The accompanying drawings are used to provide a further understanding of the embodiments of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0046] Figure 1 This is a schematic block diagram of a system for analyzing driving ability of stroke patients based on eye tracking technology according to an embodiment of the present application.

[0047] Figure 2 This is a data flow diagram of a system for analyzing driving ability of stroke patients based on eye tracking technology according to an embodiment of the present application.

[0048] Figure 3 This is a schematic block diagram of a feature filtering processing module in a stroke patient driving ability analysis system based on eye tracking technology according to an embodiment of the present application.

[0049] Figure 4 This is a schematic block diagram of a semantic contribution calculation unit in a stroke patient driving ability analysis system based on eye tracking technology according to an embodiment of the present application.

[0050] Figure 5 This is a schematic block diagram of a feature message propagation module in a stroke patient driving ability analysis system based on eye tracking technology according to an embodiment of the present application.

[0051] Figure 6 This is a schematic flowchart of a method for analyzing driving ability of stroke patients based on eye tracking technology according to an embodiment of the present application. DETAILED DESCRIPTION

[0052] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments of the present application, all other embodiments obtained by ordinary technicians in this field without creative work also fall within the scope of protection of the present application.

[0053] Eye tracking technology is an advanced technology that obtains the user's visual attention and cognitive state by monitoring and analyzing eye movements. This technology can record the individual's gaze point, gaze time, pupil changes and other information in real time, and is widely used in psychology, medicine, market research and other fields. In driving situations, eye tracking can effectively reflect the driver's attention allocation, reaction time and perception of the surrounding environment.

[0054] Based on this, in the technical solution of this application, a driving ability analysis system for stroke patients based on eye tracking technology is proposed, which can collect eye state tracking videos through eye tracking equipment, and introduce data processing and image analysis algorithms based on artificial intelligence and machine vision analysis at the back end to analyze the eye state tracking videos, so as to learn and characterize the temporal propagation aggregation representation characteristics of the eye behavior patterns of stroke patients, which are used to reflect the semantic information such as the eye gaze point position, gaze time, pupil size change of stroke patients, so as to evaluate and detect whether the patient is suitable for driving. In this way, by combining eye tracking technology with advanced artificial intelligence and deep learning algorithms, it is possible to realize intelligent analysis and objective evaluation of the driving ability of stroke patients, so as to judge whether the driver is suitable for driving and provide a scientific basis for medical professionals.

[0055] Figure 1 This is a schematic block diagram of a system for analyzing driving ability of stroke patients based on eye tracking technology according to an embodiment of the present application. Figure 2 Schematic diagram of data flow of the driving ability analysis system for stroke patients based on eye tracking technology in an embodiment of the present application. Figure 1 and Figure 2As shown, the driving ability analysis system 100 for stroke patients based on eye tracking technology includes: an eye state tracking video acquisition module 110, which is used to obtain an eye state tracking video collected by an eye tracking device; a key frame sampling module 120, which is used to perform key frame sampling on the eye state tracking video to obtain a time queue of eye state tracking key frames; an eye state feature extraction module 130, which is used to perform eye state feature extraction on each eye state tracking key frame in the time queue of the eye state tracking key frames to obtain a time queue of eye state semantic feature vectors; a feature filtering processing module 140, which is used to perform feature filtering processing based on contribution semantics on the time queue of the eye state semantic feature vectors to obtain a time queue of filtered eye state semantic feature vectors; a feature message propagation module 150, which is used to perform feature message propagation on the time queue of the filtered eye state semantic feature vectors to obtain an eye behavior pattern temporal propagation aggregation representation vector; and a driving ability analysis module 160, which is used to perform driving ability analysis based on the eye behavior pattern temporal propagation aggregation representation vector to determine whether it is suitable for driving.

[0056] In the above-mentioned stroke patient driving ability analysis system based on eye tracking technology, the eye state tracking video acquisition module 110 is used to obtain the eye state tracking video collected by the eye tracking device. It should be understood that the eye state tracking video can provide the eye movement data of the stroke patient in a specific situation, which is very important for evaluating whether the patient is suitable for driving. Eye tracking technology can capture information such as the user's gaze point, gaze duration, and pupil changes. This information helps to understand the patient's attention distribution, reaction speed, and perception of the surrounding environment. Especially for stroke patients, their cognitive functions and physical functions may be impaired, affecting their driving safety, so the data obtained by eye tracking technology can help medical professionals more objectively evaluate whether the patient has the ability to drive safely.

[0057] Optionally, in one embodiment of the present application, a special eye tracking device is used to track the video of the collected eye state. Such equipment generally includes cameras and other sensors that can record the position, direction and movement of the eyes. The collection process may be carried out in a controlled environment, such as in a simulated driving scenario, where the patient follows instructions to complete a series of tasks while the device records their eye movement data. After the data collection is completed, one or more video clips will be formed for subsequent analysis. These video clips will be input into data processing and image analysis algorithms based on artificial intelligence and machine vision analysis to further analyze the patient's eye behavior patterns and ultimately evaluate their driving ability.

[0058] In the above-mentioned stroke patient driving ability analysis system based on eye tracking technology, the key frame sampling module 120 is used to perform key frame sampling on the eye state tracking video to obtain a time queue of eye state tracking key frames. It should be understood that considering that the complete eye state tracking video contains a large amount of continuous frame data, directly processing the entire video will be very time-consuming and computationally expensive. Through key frame sampling, representative frames (i.e., key frames) can be selected from the continuous eye state tracking video, which can not only retain the main information of the video, but also reduce the amount of data for subsequent processing, improve efficiency and reduce unnecessary computing resource consumption. In addition, the selection of key frames can also help focus on those moments that best reflect the changes in eye state, thereby helping to accurately analyze the eye behavior patterns and driving abilities of stroke patients.

[0059] Optionally, in one embodiment of the present application, key frame sampling is performed on the eye state tracking video to obtain a time queue of eye state tracking key frames, including: first, determining a suitable sampling frequency or strategy, which may be based on time intervals, degree of change in eye activity, or other criteria. Then, key frames are selectively extracted from the video stream according to the selected strategy. These key frames should contain enough information to represent the content of the entire video while ensuring the efficiency of sampling. Once the key frames are selected, a time queue is constructed, which arranges all selected eye state tracking key frames in chronological order to facilitate further analysis and processing. This series of key frames will be used to extract eye state features and then evaluate the driving ability of stroke patients.

[0060] In the above-mentioned stroke patient driving ability analysis system based on eye tracking technology, the eye state feature extraction module 130 is used to extract eye state features from each eye state tracking key frame in the time queue of the eye state tracking key frame to obtain a time queue of eye state semantic feature vectors. That is, each eye state tracking key frame in the time queue of the eye state tracking key frame is respectively input into the eye state feature extractor based on the hole convolutional neural network model for feature mining, so as to respectively extract the eye state semantic feature information about the stroke patient in each eye state tracking key frame, thereby obtaining a time queue of eye state semantic feature vectors.

[0061] Optionally, in one embodiment of the present application, the eye state feature extraction module 130 is used to: input each eye state tracking key frame in the time queue of the eye state tracking key frame into an eye state feature extractor based on a void convolutional neural network model to obtain the time queue of the eye state semantic feature vector.

[0062] In the above-mentioned stroke patient driving ability analysis system based on eye tracking technology, the feature filtering processing module 140 is used to perform feature filtering processing based on contribution semantics on the time queue of the eye state semantic feature vector to obtain a filtered time queue of the eye state semantic feature vector. It should be understood that since each eye state semantic feature vector in the time queue of the eye state semantic feature vector contains semantic features related to the patient's eye state, the time queue of these vectors comprehensively reflects the eye state feature information that changes over time, such as the gaze point position, gaze time, pupil size change, etc. However, in the eye state semantic feature information presented by the eye state tracking key frames at these different time points, there will be redundant information or noise that is irrelevant to the driving ability analysis task, which will cause the risk of overfitting and affect the accuracy of subsequent tasks. Therefore, in order to extract the most representative and relevant features from the original eye state semantic feature data of key frames at a large number of time points, thereby improving the model's analysis and generalization capabilities of the patient's driving ability, in the technical solution of the present application, the time queue of the eye state semantic feature vector is further subjected to feature filtering processing based on contribution semantics to obtain a filtered time queue of the eye state semantic feature vector.

[0063] Figure 3 FIG. 1 is a schematic block diagram of a feature filtering processing module in a stroke patient driving ability analysis system based on eye tracking technology in an embodiment of the present application. Figure 3 As shown, optionally, in one embodiment of the present application, the feature filtering processing module 140 includes: a cluster analysis unit 141, which is used to perform cluster analysis on the time queue of the eye state semantic feature vector to obtain an eye state semantic cluster center representation vector; a semantic contribution calculation unit 142, which is used to calculate the semantic contribution of each eye state semantic feature vector in the time queue of the eye state semantic feature vector relative to the eye state semantic cluster center representation vector to obtain a sequence of masked sparse eye state semantic contribution representation matrices; a feature filtering calculation unit 143, which is used to use each masked sparse eye state semantic contribution representation matrix in the sequence of masked sparse eye state semantic contribution representation matrices as a modulation matrix, and respectively calculate the matrix product between it and each eye state semantic feature vector to obtain the time queue of the filtered eye state semantic feature vector.

[0064] Optionally, in one embodiment of the present application, the cluster analysis unit 141 is used to: calculate the positional mean between each eye state semantic feature vector in the time queue of the eye state semantic feature vector to obtain the eye state semantic cluster center representation vector.

[0065] Figure 4Schematic block diagram of a semantic contribution calculation unit in the driving ability analysis system for stroke patients based on eye tracking technology in an embodiment of the present application. Figure 4 As shown, optionally, in one embodiment of the present application, the semantic contribution calculation unit 142 includes: a semantic contribution representation matrix calculation subunit 1421, which is used to use the eye state semantic clustering center representation vector as a pseudo anchor center, and calculate the semantic contribution representation matrix of each eye state semantic feature vector in the time queue of the eye state semantic feature vector relative to the pseudo anchor center to obtain a sequence of eye state semantic contribution representation matrices; a random inactivation processing subunit 1422, which is used to perform random inactivation processing on each eye state semantic contribution representation matrix in the sequence of eye state semantic contribution representation matrices to obtain a sequence of masked sparse eye state semantic contribution representation matrices.

[0066] Optionally, in one embodiment of the present application, the semantic contribution representation matrix calculation subunit 1421 is used to: calculate the positional division between the transposed vector of the eye state semantic cluster center representation vector and the bi-norm of the eye state semantic cluster center representation vector to obtain the transformed eye state semantic cluster center vector; calculate the vector multiplication between each eye state semantic feature vector and the transformed eye state semantic cluster center vector, and use The function performs soft maximum normalization processing on the obtained sequence of association matrices to obtain the sequence of eye state semantic contribution representation matrices.

[0067] In summary, in the technical solution of the present application, the following feature filtering formula performs feature filtering processing based on contribution semantics on the time queue of the eye state semantic feature vector to obtain the filtered time queue of the eye state semantic feature vector;

[0068] Among them, the feature filtering formula is:

[0069] ;

[0070] ;

[0071] ;

[0072] ;

[0073] in, is the time queue of the eye state semantic feature vector, are the first, second and third in the time queue of the eye state semantic feature vector respectively. eye state semantic feature vector, is the first eye state semantic feature vector, is the number of feature vectors in the time queue of the eye state semantic feature vector, is the eye state semantic cluster center representation vector, represents the two-norm of a vector, for function, For random dropout, For the The masked sparse eye state semantic contribution representation matrix corresponding to the eye state semantic feature vectors is: is the first A filtered eye state semantic feature vector.

[0074] Specifically, the feature filtering process based on contribution semantics includes the following steps: first, clustering analysis is performed on the time queue of the eye state semantic feature vector to obtain the eye state semantic cluster center representation vector as a pseudo anchor center, which helps to capture the intrinsic structure and distribution of the eye state semantic feature time series data. The determination of the pseudo anchor center can be used as a reference point to evaluate the importance of each eye state semantic feature vector. This ensures that the model focuses on those features that are closely related to the driving behavior pattern, thereby improving the quality of feature representation, and helps to more accurately provide a reference point for driving ability analysis and detection tasks, so as to strengthen these features and eye state semantic information related to subsequent tasks. Next, the semantic contribution of each eye state semantic feature vector relative to the eye state semantic cluster center representation vector is calculated, and a sequence of masked sparse eye state semantic contribution representation matrices is generated to quantify the contribution and relative importance of each eye state semantic feature to the overall eye state semantics, providing a basis for feature screening and enhanced expression. Furthermore, the random dropout mechanism is used to sparse the semantic contribution representation matrix of each eye state, and a sequence of masked sparse eye state semantic contribution representation matrices is generated to enhance the generalization ability of the model and reduce the number of eye state semantic features. Specifically, by randomly "closing" some unimportant features, the model's attention to the key eye state temporal semantic features is strengthened, while the complexity of the model is reduced. Finally, the masked sparse eye state semantic contribution representation matrix is ​​used as a modulation matrix to perform matrix product operations with each eye state semantic feature vector to achieve feature filtering, thereby filtering out unimportant or noise features and retaining the information that is most helpful for subsequent driving ability detection tasks. In this way, it can be ensured that the data finally input into the driving ability analysis module is the most representative and informative, so that the system can more accurately assess whether stroke patients are suitable for driving. This method not only improves the detection performance of the system, but also enables the model to maintain good stability and adaptability in different driving scenarios.

[0075] In the above-mentioned stroke patient driving ability analysis system based on eye tracking technology, the feature message propagation module 150 is used to perform feature message propagation on the time queue of the filtered eye state semantic feature vector to obtain the eye behavior pattern time series propagation aggregation representation vector. It should be understood that since each filtered eye state semantic feature vector in the time queue of the filtered eye state semantic feature vector respectively contains the patient's eye state semantic features after feature enhancement and filtering, and these eye state semantic features have correlation relationships and interactive connections in the time dimension, this connection can be used to reflect the patient's eye gaze point position movement, gaze time, pupil size changes, etc. In addition, it is also considered that in the time dimension, the eye semantics at different time points will have different degrees of influence on the eye semantics at the current time point, and this degree of influence has a spatiotemporal transmission characteristic. Therefore, in order to capture the dynamic changes of eye behavior patterns from time series data, and integrate this information to form a time-series propagation aggregate representation vector that can reflect the eye behavior pattern to more accurately analyze driving ability, in the technical solution of the present application, the time queue of the filtered eye state semantic feature vector is further subjected to feature message propagation to obtain the eye behavior pattern time-series propagation aggregate representation vector. It should be understood that the change of eye state is a dynamic process that occurs over time. The feature message propagation network can effectively capture the temporal evolution of eye behavior patterns by considering the spatiotemporal relationship between nodes. This means that not only is the eye state at the current moment important, but the past state also provides useful contextual information. By comprehensively considering this information, more accurate driving ability analysis results can be obtained.

[0076] Figure 5 FIG. 1 is a schematic block diagram of a feature message propagation module in a stroke patient driving ability analysis system based on eye tracking technology according to an embodiment of the present application. Figure 5As shown, optionally, in one embodiment of the present application, the feature message propagation module 150 includes: a current filtered feature vector extraction unit 151, which is used to extract the current filtered eye state semantic feature vector from the time queue of the filtered eye state semantic feature vector, and define other filtered eye state semantic feature vectors in the time queue of the filtered eye state semantic feature vector as historical filtered eye state semantic feature vectors to obtain the time queue of historical filtered eye state semantic feature vectors; an eye state semantic static energy value calculation unit 152, which is used to calculate the time queue of the historical filtered eye state semantic feature vectors. The eye state semantic static energy value of each historical filtered eye state semantic feature vector in the queue; an eye state semantic energy spatial attenuation factor determination unit 153, used to determine the eye state semantic energy spatial attenuation factor based on the spatial span coefficient between the eye state semantic feature vector after each historical filter and the eye state semantic feature vector after the current filter; an eye state semantic energy time attenuation factor determination unit 154, used to determine the eye state semantic energy time attenuation factor based on the time span coefficient between the eye state semantic feature vector after each historical filter and the eye state semantic feature vector after the current filter; an eye state semantic energy dynamic attenuation factor determination unit 154, used to determine the eye state semantic energy time attenuation factor based on the time span coefficient between the eye state semantic feature vector after each historical filter and the eye state semantic feature vector after the current filter; The eye state semantic energy spatiotemporal transfer value calculation unit 155 is used to perform spatiotemporal modulation on the eye state semantic static energy value of each of the historical filtered eye state semantic feature vectors based on the eye state semantic energy spatial attenuation factor and the eye state semantic energy temporal attenuation factor to obtain the eye state semantic dynamic energy spatiotemporal transfer value of each of the historical filtered eye state semantic feature vectors relative to the currently filtered eye state semantic feature vector; the eye state semantic dynamic energy spatiotemporal transfer weight calculation unit 156 is used to input the eye state semantic dynamic energy spatiotemporal transfer value of each of the historical filtered eye state semantic feature vectors into a mask function based on the mask function. An energy transfer gating module is used to obtain a sequence of eye state semantic dynamic energy spatiotemporal transfer weights; a transfer aggregation modulation vector calculation unit 157 is used to calculate the weighted sum of the time queue of the historically filtered eye state semantic feature vectors based on the sequence of eye state semantic dynamic energy spatiotemporal transfer weights to obtain the historically filtered eye state semantic node feature transfer aggregation modulation vector; a pattern temporal propagation aggregation representation vector calculation unit 158 ​​is used to calculate the positional sum of the historically filtered eye state semantic node feature transfer aggregation modulation vector and the currently filtered eye state semantic feature vector to obtain the eye behavior pattern temporal propagation aggregation representation vector.

[0077] Optionally, in one embodiment of the present application, the eye state semantic energy space attenuation factor determination unit 153 is used to: calculate the spatial distance span between the eye state semantic feature vector after historical filtering and the eye state semantic feature vector after current filtering to obtain the eye state semantic space span coefficient after historical filtering; calculate the division between the eye state semantic static energy value corresponding to the eye state semantic feature vector after historical filtering and the eye state semantic space span coefficient after historical filtering to obtain the eye state semantic energy space attenuation factor.

[0078] Optionally, in one embodiment of the present application, the eye state semantic static energy value calculation unit 152 is used to: respectively calculate the mean and standard deviation of the historically filtered eye state semantic feature vector to obtain the historically filtered eye state semantic feature mean and the historically filtered eye state semantic standard deviation; calculate the fourth power of the position difference between the historically filtered eye state semantic feature vector and the historically filtered eye state semantic feature mean, and then calculate the expected value of the fourth power modulated historically filtered eye state semantic offset vector to obtain the filtered eye state semantic feature offset expected factor; calculate the division between the filtered eye state semantic feature offset expected factor and the fourth power of the historically filtered eye state semantic standard deviation, and then add the calculated value to the maximum eigenvalue in the historically filtered eye state semantic feature vector to obtain the eye state semantic static energy value.

[0079] Optionally, in one embodiment of the present application, the eye state semantic energy time decay factor determination unit 154 is used to: calculate the subtraction between the timestamp of the currently filtered eye state semantic feature vector and the timestamp of the historically filtered eye state semantic feature vector, and then round down the obtained timestamp value to obtain the filtered eye state semantic time span coefficient; divide the filtered eye state semantic time span coefficient by the inverse scaling parameter that controls the time decay period, and then use the obtained time span decay representation coefficient as an exponential power to calculate the natural exponential function value with the natural constant e as the base to obtain the time span decay class support representation coefficient; calculate the division between the eye state semantic static energy value and the time span decay class support representation coefficient to obtain the eye state semantic energy time decay factor.

[0080] Optionally, in one embodiment of the present application, the eye state semantic dynamic energy spatiotemporal transfer weight calculation unit 156 is used to: in the energy transfer gating module based on the mask function, in response to the eye state semantic dynamic energy spatiotemporal transfer value being greater than a threshold hyperparameter, The function normalizes the eye state semantic dynamic energy spatiotemporal transfer value that is greater than the threshold hyperparameter to obtain the eye state semantic dynamic energy spatiotemporal transfer weight.

[0081] In summary, in the embodiment of the present application, the following characteristic message propagation formula is used to perform characteristic message propagation on the time queue of the filtered eye state semantic feature vector to obtain the eye behavior pattern time series propagation aggregation representation vector;

[0082] Among them, the characteristic message propagation formula is:

[0083] ;

[0084] ;

[0085] ;

[0086] ;

[0087] ;

[0088] ;

[0089] ;

[0090] in, is the time queue of the filtered eye state semantic feature vector, are the first, second, and third in the time queue of the filtered eye state semantic feature vector. and The filtered eye state semantic feature vector, For the The first of the filtered eye state semantic feature vectors The eigenvalues ​​at the positions, and They are respectively The feature mean and standard deviation of the filtered eye state semantic feature vector, is to calculate the expected value, It means to take the maximum value in the vector. is the semantic static energy value of the eye state, express and The spatial span coefficient between for and The energy space attenuation factor between and express and timestamp, Indicates a round-down operation. is the descaling parameter that controls the time decay period, is the value of the natural exponential function, for and The energy time decay factor between and is a trainable hyperparameter, for Relative to The dynamic energy space-time transfer value of For mask operation, is the number of vectors in the time queue of the filtered eye state semantic feature vector, is the threshold hyperparameter, for function, A representation vector is aggregated for the temporal propagation of the eye behavior pattern.

[0091] Specifically, the step of propagating the feature message includes: first, extracting the current filtered eye state semantic feature vector and the time queue of the historical filtered eye state semantic feature vector from the time queue of the filtered eye state semantic feature vector. Then, calculating the static energy value of each historical filtered eye state semantic feature vector in the time queue of the historical filtered eye state semantic feature vector, the static energy value is related to the maximum eigenvalue, mean and variance of each historical filtered eye state semantic feature vector, the static energy value provides a quantitative measure of the inherent properties of the historical filtered eye state semantic feature vector in space, provides a benchmark for the spatiotemporal attenuation of potential energy, and helps the model understand the static importance of the eye state semantics at each time node in the spatial dimension and the temporal dimension. Then, based on the spatial span coefficient between each historical filtered eye state semantic feature vector and the current filtered eye state semantic feature vector, the eye state semantic energy spatial attenuation factor is determined, and based on the time span coefficient between each historical filtered eye state semantic feature vector and the current filtered eye state semantic feature vector, the eye state semantic energy temporal attenuation factor is determined. The spatial span coefficient and the temporal span coefficient reflect the spatial distance and temporal distance between time nodes respectively. The attenuation factor can adjust the weight of information transmission between nodes based on these coefficients, that is, pay attention to more important semantic information related to the subsequent driving ability analysis task in the process of temporal information propagation of eye state semantic features. Through the adjustment of the energy decay factor, the model can capture locality and timeliness characteristics more sensitively. This means that the model can distinguish which information is most important under specific time and space conditions, so that when processing eye state changes, it can more accurately identify key behavior patterns. Furthermore, based on the eye state semantic energy spatial decay factor and the eye state semantic energy temporal decay factor, the eye state semantic static energy value of each historical filtered eye state semantic feature vector is spatiotemporally modulated, so as to adjust the static energy value by combining the eye state semantic energy spatial decay factor and the eye state semantic energy temporal decay factor through spatiotemporal modulation to reflect the dynamic changes of the node eye state semantic features in time and space. It should be understood that the effect of spatiotemporal modulation is to make the information propagation process about the eye state semantics in the node features more consistent with the actual physical process, strengthen the model's ability to capture the dynamic changes of the eye state time series, and improve the model's processing effect on time-sensitive tasks. Further, the eye state semantic dynamic energy spatiotemporal transfer value of each historically filtered eye state semantic feature vector is input into the energy transfer gating module based on the mask function. The mask function is used to filter and strengthen important eye state semantic spatiotemporal transfer information. The gating mechanism determines which eye state information is important and which information can be ignored through learning.In this way, the accuracy and effectiveness of information transfer are improved by energy transfer gating, and by adaptively adjusting the information flow, the model can focus more on the features that contribute to the subsequent driving ability analysis and detection tasks. Then, based on the sequence of the eye state semantic dynamic energy spatiotemporal transfer weights, the weighted sum of the time queue of the eye state semantic feature vector after historical filtering is calculated to obtain the eye state semantic node feature transfer aggregation modulation vector after historical filtering. That is, the eye state semantic feature information from different historical time nodes is integrated by weighted summing of the dynamic energy spatiotemporal transfer values ​​of the eye state semantic feature vector after historical filtering, so that the model can use the contextual information in the time series and improve the richness and accuracy of the eye state semantic feature aggregation representation in the historical node. Finally, the eye state semantic node feature transfer aggregation modulation vector after historical filtering and the eye state semantic feature vector after current filtering are summed by position to generate an eye behavior pattern temporal propagation aggregation representation vector that integrates the historical eye state semantic aggregation information and the current eye state semantics. The temporal propagation aggregation representation vector of the eye behavior pattern contains feature representations with rich contextual semantics. It not only includes the semantic feature information of the current eye state, but also integrates the information propagation influence of the previous eye state, thereby providing more comprehensive and accurate data support for subsequent driving ability analysis and enhancing the model's processing and generalization capabilities for complex tasks.

[0092] In the above-mentioned stroke patient driving ability analysis system based on eye tracking technology, the driving ability analysis module 160 is used to perform driving ability analysis based on the eye behavior pattern time series propagation aggregation representation vector to determine whether it is suitable for driving. Optionally, in one embodiment of the present application, the driving ability analysis module 160 is used to: input the eye behavior pattern time series propagation aggregation representation vector into the driving ability analysis module based on the classifier to obtain an analysis result, and the analysis result is used to indicate whether it is suitable for driving. That is to say, by using the time series information propagation aggregation representation characteristics of the eye behavior pattern to perform classification processing, the patient's driving ability is analyzed, so as to detect the eye gaze point position, gaze time, pupil size change, etc. of the stroke patient, so as to evaluate and detect whether the patient is suitable for driving. In this way, by combining eye tracking technology with advanced artificial intelligence and deep learning algorithms, it is possible to realize intelligent analysis and objective evaluation of the driving ability of stroke patients, so as to judge whether the driver is suitable for driving, and provide a scientific basis for medical professionals.

[0093] In a preferred example, inputting the eyeball behavior pattern temporal propagation aggregation representation vector into a classifier-based driving ability analysis module to obtain an analysis result includes:

[0094] Determine the maximum eigenvalue of the eyeball behavior pattern time series propagation aggregation and the minimum eigenvalue of the eyeball behavior pattern time series propagation aggregation of the eyeball behavior pattern time series propagation aggregation representation vector, and calculate the eyeball behavior pattern time series propagation aggregation mean and the eyeball behavior pattern time series propagation aggregation standard deviation of the feature set of the eyeball behavior pattern time series propagation aggregation representation vector;

[0095] Calculate the quotient of the aggregated mean of the eye behavior pattern time series propagation and the aggregated standard deviation of the eye behavior pattern time series propagation to obtain an aggregated statistical normalization value of the eye behavior pattern time series propagation;

[0096] Calculate the inverse of each eigenvalue of the eye behavior pattern time series propagation aggregation representation vector, multiply it by the difference between the maximum eigenvalue of the eye behavior pattern time series propagation aggregation and the minimum eigenvalue of the eye behavior pattern time series propagation aggregation, and then subtract it from the statistical normalization value of the eye behavior pattern time series propagation aggregation to obtain an eye behavior pattern time series propagation aggregation distribution approximation vector;

[0097] Calculate an exponential function with a natural constant as the base, with each eigenvalue of the eyeball behavior pattern time series propagation aggregate distribution approximation vector as the exponent, to obtain an eyeball behavior pattern time series propagation aggregate distribution class approximation vector;

[0098] Performing point-wise addition of the eyeball behavior pattern time series propagation aggregation distribution class approximation vector and the eyeball behavior pattern time series propagation aggregation statistical normalization value, and calculating the base-two logarithm of the absolute value of each eigenvalue of the point-wise addition vector to obtain an optimized eyeball behavior pattern time series propagation aggregation representation vector;

[0099] The optimized eye behavior pattern temporal propagation aggregation representation vector is input into a classifier-based driving ability analysis module to obtain an analysis result.

[0100] The optimized representation of the eyeball behavior pattern temporal propagation aggregation representation vector is:

[0101] ;

[0102] ;

[0103] ;

[0104] ;

[0105] in, Indicates the maximum eigenvalue of the temporal propagation aggregation of the eyeball behavior pattern, Indicates the minimum eigenvalue of the temporal propagation aggregation of eye behavior patterns, represents the aggregate mean of the temporal propagation of eye behavior patterns, represents the aggregate standard deviation of the temporal propagation of eye behavior patterns, Aggregate statistical normalization values ​​for temporal propagation of eye behavior patterns, and is the difference between the maximum eigenvalue of the eyeball behavior pattern temporal propagation aggregation and the minimum eigenvalue of the eyeball behavior pattern temporal propagation aggregation, represents the temporal propagation aggregation representation vector of the eyeball behavior pattern, It means to calculate the inverse of each eigenvalue of the vector. It means point multiplication by position. It means adding by position. It means subtracting by position. Represents the approximate vector of the temporal propagation aggregation distribution of eyeball behavior patterns, Representation of optimized eye behavior pattern temporal propagation aggregate representation vector.

[0106] Here, in the preferred example, since the time queues of the filtered eye state semantic feature vectors respectively represent the filtered query semantic features of the eye state tracking key frames of each time node based on the pseudo-center contribution of the embedded image semantic features, when the feature message is propagated, the node semantic filtering query differences of the eye state will have different semantic propagation weights based on the potential energy spatiotemporal transfer characteristics, so that the eye behavior pattern temporal propagation aggregation representation vector will also have a diversified set expression distribution of temporal aggregation features. Therefore, it is expected to improve the balance between the regression mapping accuracy and completeness when the eye behavior pattern temporal propagation aggregation representation vector is input into the driving ability analysis module based on the classifier for class regression, thereby improving the accuracy of the analysis results.

[0107] Therefore, by performing random statistical normalization on the diversified feature set of the eye behavior pattern time series propagation aggregate representation vector, the confidence space constructed based on the overall feature value of the eye behavior pattern time series propagation aggregate representation vector is approximated by the standardized continuous probability density distribution relative to the response hypothesis test of each feature value of the eye behavior pattern time series propagation aggregate representation vector, thereby establishing the target reachability of the diversified feature distribution of the eye behavior pattern time series propagation aggregate representation vector to the unified regression target, so as to achieve the balanced executableness between mapping accuracy and mapping completeness in the class regression process based on the diversified feature distribution of the eye behavior pattern time series propagation aggregate representation vector, and improve the accuracy of the analysis results obtained by the eye behavior pattern time series propagation aggregate representation vector input into the driving ability analysis module based on the classifier. In this way, the patient's driving ability can be analyzed more accurately, so as to objectively evaluate and detect whether the stroke patient is suitable for driving, so as to judge whether the driver is suitable for driving, and provide a scientific basis for medical professionals.

[0108] In summary, according to the embodiment of the present application, the driving ability analysis system for stroke patients based on eye tracking technology is explained, which can collect eye state tracking video through eye tracking equipment, and introduce data processing and image analysis algorithms based on artificial intelligence and machine vision analysis at the back end to analyze the eye state tracking video, so as to learn and characterize the temporal propagation aggregation representation characteristics of the eye behavior pattern of stroke patients, which is used to reflect the semantic information such as the eye gaze point position, gaze time, pupil size change of stroke patients, so as to evaluate and detect whether the patient is suitable for driving. In this way, by combining eye tracking technology with advanced artificial intelligence and deep learning algorithms, it is possible to realize intelligent analysis and objective evaluation of the driving ability of stroke patients, so as to judge whether the driver is suitable for driving, and provide a scientific basis for medical professionals.

[0109] Figure 6 Schematic flow chart of the method for analyzing driving ability of stroke patients based on eye tracking technology according to an embodiment of the present application. Figure 6 As shown, a method for analyzing driving ability of stroke patients based on eye tracking technology includes: S1, obtaining an eye state tracking video collected by an eye tracking device; S2, performing key frame sampling on the eye state tracking video to obtain a time queue of eye state tracking key frames; S3, performing eye state feature extraction on each eye state tracking key frame in the time queue of the eye state tracking key frames to obtain a time queue of eye state semantic feature vectors; S4, performing feature filtering processing based on contribution semantics on the time queue of the eye state semantic feature vectors to obtain a time queue of filtered eye state semantic feature vectors; S5, performing feature message propagation on the time queue of the filtered eye state semantic feature vectors to obtain an eye behavior pattern temporal propagation aggregation representation vector; S6, performing driving ability analysis based on the eye behavior pattern temporal propagation aggregation representation vector to determine whether it is suitable for driving.

[0110] Here, those skilled in the art can understand that the specific operations of each step in the above-mentioned method for analyzing driving ability of stroke patients based on eye tracking technology have been described in the above reference. Figures 1 to 5 The driving ability analysis system for stroke patients based on eye tracking technology has been introduced in detail, and therefore, its repeated description will be omitted.

[0111] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference to a figure in a claim should not be considered as limiting the claim to which it relates.

[0112] Finally, it should be noted that the above description has been given for the purpose of illustration and description. In addition, the above embodiments are only used to illustrate the technical solution of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention.

Claims

1. A driving ability analysis system for stroke patients based on eye tracking technology, characterized in that: include: An eyeball state tracking video acquisition module is used to acquire the eyeball state tracking video acquired by the eyeball tracking device; A key frame sampling module, used for performing key frame sampling on the eye state tracking video to obtain a time queue of eye state tracking key frames; An eye state feature extraction module, used for extracting eye state features from each eye state tracking key frame in the time queue of the eye state tracking key frames to obtain a time queue of eye state semantic feature vectors; A feature filtering processing module, used for performing feature filtering processing based on contribution semantics on the time queue of the eyeball state semantic feature vector to obtain a filtered time queue of the eyeball state semantic feature vector; Wherein, the feature filtering processing module includes: a cluster analysis unit, which is used to perform cluster analysis on the time queue of the eye state semantic feature vector to obtain an eye state semantic cluster center representation vector; a semantic contribution calculation unit, which is used to calculate the semantic contribution of each eye state semantic feature vector in the time queue of the eye state semantic feature vector relative to the eye state semantic cluster center representation vector to obtain a sequence of masked sparse eye state semantic contribution representation matrices; a feature filtering calculation unit, which is used to use each masked sparse eye state semantic contribution representation matrix in the sequence of masked sparse eye state semantic contribution representation matrices as a modulation matrix, and respectively calculate the matrix product between it and each eye state semantic feature vector to obtain the time queue of the filtered eye state semantic feature vector; A feature message propagation module, used for performing feature message propagation on the time queue of the filtered eye state semantic feature vector to obtain an eye behavior pattern time series propagation aggregation representation vector; The driving ability analysis module is used to perform driving ability analysis based on the eye behavior pattern temporal propagation aggregation representation vector to determine whether the driver is suitable for driving.

2. The driving ability analysis system for stroke patients based on eye tracking technology according to claim 1 is characterized in that: The eye state feature extraction module is used to: input each eye state tracking key frame in the time queue of the eye state tracking key frame into an eye state feature extractor based on a void convolutional neural network model to obtain the time queue of the eye state semantic feature vector.

3. The driving ability analysis system for stroke patients based on eye tracking technology according to claim 2 is characterized in that: The cluster analysis unit is used to calculate the positional mean between each eye state semantic feature vector in the time queue of the eye state semantic feature vector to obtain the eye state semantic cluster center representation vector.

4. The driving ability analysis system for stroke patients based on eye tracking technology according to claim 3 is characterized in that: The semantic contribution calculation unit comprises: A semantic contribution representation matrix calculation subunit is used to use the eye state semantic cluster center representation vector as a pseudo anchor center, and calculate the semantic contribution representation matrix of each eye state semantic feature vector in the time queue of the eye state semantic feature vector relative to the pseudo anchor center to obtain a sequence of eye state semantic contribution representation matrices; The random deactivation processing subunit is used to perform random deactivation processing on each eye state semantic contribution representation matrix in the sequence of eye state semantic contribution representation matrices to obtain the sequence of masked sparse eye state semantic contribution representation matrices.

5. The driving ability analysis system for stroke patients based on eye tracking technology according to claim 4 is characterized in that: The semantic contribution representation matrix calculation subunit is used to: Calculating the positional division between the transposed vector of the eye state semantic cluster center representation vector and the bi-norm of the eye state semantic cluster center representation vector to obtain a transformed eye state semantic cluster center vector; After calculating the vector multiplication between each eye state semantic feature vector and the transformed eye state semantic cluster center vector, use The function performs soft maximum normalization processing on the obtained sequence of association matrices to obtain the sequence of eye state semantic contribution representation matrices.

6. The driving ability analysis system for stroke patients based on eye tracking technology according to claim 5 is characterized in that: The feature message propagation module comprises: a current filtered feature vector extraction unit, configured to extract a current filtered eye state semantic feature vector from the time queue of the filtered eye state semantic feature vectors, and define other filtered eye state semantic feature vectors in the time queue of the filtered eye state semantic feature vectors as historical filtered eye state semantic feature vectors to obtain a time queue of historical filtered eye state semantic feature vectors; An eye state semantic static energy value calculation unit, used for calculating the eye state semantic static energy value of each historical filtered eye state semantic feature vector in the time queue of the historical filtered eye state semantic feature vector; An eye state semantic energy spatial attenuation factor determination unit, used to determine the eye state semantic energy spatial attenuation factor based on the spatial span coefficient between the eye state semantic feature vectors after each historical filtering and the eye state semantic feature vector after the current filtering; An eye state semantic energy time decay factor determination unit, used to determine the eye state semantic energy time decay factor based on the time span coefficient between the eye state semantic feature vectors after each historical filtering and the eye state semantic feature vector after the current filtering; An eye state semantic dynamic energy spatiotemporal transfer value calculation unit is used to perform spatiotemporal modulation on the eye state semantic static energy value of each of the historically filtered eye state semantic feature vectors based on the eye state semantic energy spatial attenuation factor and the eye state semantic energy temporal attenuation factor to obtain the eye state semantic dynamic energy spatiotemporal transfer value of each of the historically filtered eye state semantic feature vectors relative to the currently filtered eye state semantic feature vector; An eye state semantic dynamic energy spatiotemporal transfer weight calculation unit is used to input the eye state semantic dynamic energy spatiotemporal transfer value of each of the historically filtered eye state semantic feature vectors into an energy transfer gating module based on a mask function to obtain a sequence of eye state semantic dynamic energy spatiotemporal transfer weights; A transfer aggregation modulation vector calculation unit is used to calculate the weighted sum of the time queue of the eye state semantic feature vector after historical filtering based on the sequence of the eye state semantic dynamic energy spatiotemporal transfer weights to obtain the eye state semantic node feature transfer aggregation modulation vector after historical filtering; A pattern time series propagation aggregate representation vector calculation unit, used for calculating the positional sum of the eye state semantic node feature transfer aggregate modulation vector after historical filtering and the eye state semantic feature vector after current filtering to obtain the eye behavior pattern time series propagation aggregate representation vector; Wherein, the eye state semantic energy space attenuation factor determination unit is used to: Calculating the spatial distance span between the eye state semantic feature vector after historical filtering and the eye state semantic feature vector after current filtering to obtain the eye state semantic spatial span coefficient after historical filtering; The eye state semantic static energy value corresponding to the eye state semantic feature vector after historical filtering is divided by the eye state semantic space span coefficient after historical filtering to obtain the eye state semantic energy space attenuation factor.

7. The driving ability analysis system for stroke patients based on eye tracking technology according to claim 6 is characterized in that: The eyeball state semantic static energy value calculation unit is used to: Calculating the mean and standard deviation of the eye state semantic feature vector after historical filtering to obtain the mean of the eye state semantic feature after historical filtering and the standard deviation of the eye state semantic feature after historical filtering; After calculating the fourth power of the position difference between the historical filtered eye state semantic feature vector and the historical filtered eye state semantic feature mean, the fourth power is calculated to modulate the expected value of the historical filtered eye state semantic offset vector to obtain the filtered eye state semantic feature offset expected factor; After calculating the division between the expected factor of the filtered eye state semantic feature offset and the fourth power of the historical filtered eye state semantic standard deviation, it is added to the maximum eigenvalue in the historical filtered eye state semantic feature vector to obtain the eye state semantic static energy value.

8. The driving ability analysis system for stroke patients based on eye tracking technology according to claim 7 is characterized in that: The eye state semantic energy time decay factor determination unit is used to: After calculating the subtraction between the timestamp of the currently filtered eye state semantic feature vector and the timestamp of the historically filtered eye state semantic feature vector, the obtained timestamp value is rounded down to obtain a filtered eye state semantic time span coefficient; After dividing the filtered eye state semantic time span coefficient by the inverse scaling parameter controlling the time decay period, the obtained time span decay representation coefficient is used as an exponential power, and a natural exponential function value with a natural constant e as the base is calculated to obtain a time span decay class support representation coefficient; The division between the eye state semantic static energy value and the time span attenuation class support representation coefficient is calculated to obtain the eye state semantic energy time attenuation factor.

9. The driving ability analysis system for stroke patients based on eye tracking technology according to claim 8 is characterized in that: The eyeball state semantic dynamic energy spatiotemporal transfer weight calculation unit is used to: in the energy transfer gating module based on the mask function, in response to the eyeball state semantic dynamic energy spatiotemporal transfer value being greater than a threshold hyperparameter, The function normalizes the eye state semantic dynamic energy spatiotemporal transfer value that is greater than the threshold hyperparameter to obtain the eye state semantic dynamic energy spatiotemporal transfer weight.

10. The driving ability analysis system for stroke patients based on eye tracking technology according to claim 9 is characterized in that: The driving ability analysis module is used to: input the eye behavior pattern time series propagation aggregation representation vector into the classifier-based driving ability analysis module to obtain an analysis result, and the analysis result is used to indicate whether the person is suitable for driving.

Images