An acceleration-based personalized virtual reality black insertion method for the disabled

Abstract

The application discloses an acceleration-based personalized virtual reality black insertion method for the disabled, which comprises the following steps: step 1, constructing an acceleration-based and personalized black insertion duration dataset based on the types of the disabled users; wherein the acceleration is divided into T types according to the size, and the disabled users have N types; step 2, obtaining the acceleration of the user, and calculating the time domain features and the frequency domain features of the acceleration; step 3, constructing a probability neural network based on the time domain and frequency domain features of the acceleration, inputting the time domain features and the frequency domain features of the acceleration into the probability neural network, and classifying the acceleration; and step 4, judging the type of the disabled user, combining the acceleration classification obtained in step 3, and obtaining the personalized black insertion duration of the user through the personalized black insertion duration dataset. The probability neural network based on the time domain and frequency domain features of the acceleration can well reflect the differences between different accelerations, reduce the influence of abnormal values on the network accuracy, and improve the differences between different features.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of display control, and particularly to an acceleration-based personalized virtual reality black insertion method for the disabled. BACKGROUND

[0002] Currently, virtual reality devices can be applied in various industries, which means that the application population of virtual reality devices is more extensive. Users can control existing virtual reality devices using their eyes, limbs, mouths, etc. Virtual reality devices complete target tasks according to instructions. The above multiple instruction delivery methods make it possible for the disabled to use virtual reality devices. Since the frame rate of the virtual reality device picture is fixed, the current black insertion method is commonly used to improve the smoothness and clarity of the picture. The existing black insertion technology inserts a full black picture between every two frames of video, which not only increases the frame rate of the virtual display device by 1 times, but also, due to the human eye's visual persistence phenomenon, the virtual reality display device after black insertion solves the picture trailing problem to some extent and improves the smoothness of the picture. However, the existing virtual reality black insertion technology is not friendly to the disabled. For people with weak vision, they can capture very little picture information per second, and too dense pictures may cause discomfort to people with weak vision, and such people may need a relatively longer black insertion display time; for people with physical disabilities, the picture playback speed cannot be controlled by themselves, and such people have poor mobility, and slight acceleration may contain a strong requirement for changing the picture experience; for deaf people, their attention is mainly focused on the picture, and they require higher picture smoothness, and such people may need a relatively shorter black insertion display time.

[0003] In the existing virtual reality black insertion technology, the patent application with the application number 202111473884.9 proposes to preset an acceleration and black insertion control parameter mapping table in the virtual reality device. During the use of the device, when the device acceleration changes, the preset acceleration and black insertion control parameter mapping table is queried to determine the black insertion time. However, this method has three problems: (1) The patent does not introduce the specific identification method of acceleration. (2) The patent does not introduce the method of making the acceleration and black insertion control parameter mapping table. (3) The user types are not divided in detail, and detailed black insertion strategies are not formulated according to the user types. If there is a virtual reality black insertion technology that accurately identifies the device acceleration and formulates different black insertion time data sets for different types of users, it can provide different picture experiences for different types of disabled people, so that the disabled can experience more happiness in the virtual reality device.

[0004] The traditional acceleration feature extraction method has the following problems: (1) directly extracting the acceleration sequence feature will have the problem that the difference between different acceleration features is not large; (2) due to the interference of environmental factors, there are many abnormal points in the acceleration sequence, and the traditional acceleration feature extraction cannot solve the problem of many abnormal points; (3) the correlation between the acceleration feature components is not learned, which causes the network recognition accuracy to be unable to improve. SUMMARY

[0005] In order to solve the above problems, the application provides a virtual reality black insertion method for personalized data analysis of obstacle groups based on acceleration, which comprises a probability neural network based on acceleration time domain and frequency domain features, extracts the time domain and frequency domain features of acceleration as the input of the network, and the time domain and frequency domain features of acceleration can not be disturbed by abnormal values when acting together, can well reflect the difference between different accelerations, reduce the influence of abnormal values on network accuracy, improve the difference between different features, and thus improve network accuracy.

[0006] Then, the Mahalanobis distance is used instead of the Euclidean distance to calculate the correlation of each feature component of the acceleration; the Mahalanobis distance does not need to be normalized because it is not affected by the data measurement unit, and the Mahalanobis distance can calculate the correlation of each feature component, and the application uses the Mahalanobis distance instead of the Euclidean distance in the probability neural network, which can well calculate the distance between the test sample and the training sample and improve the network recognition accuracy, and at the same time, the normalization processing is not needed, which improves the network calculation speed.

[0007] Finally, the application proposes a segmented data set updating strategy to construct a 63-class data set, after determining the user type, the personalized black insertion display setting of a specific user is realized according to the user type and the user acceleration, and the different needs of all types of users are met.

[0008] The application provides a personalized virtual reality black insertion method for obstacle groups based on acceleration, which comprises the following steps:

[0009] Step 1, constructing a personalized black insertion duration data set based on acceleration and obstacle user types; wherein the acceleration is divided into T types according to size, and the obstacle users have N types;

[0010] The black insertion duration data set of different types of obstacle users is represented as {A1, A2, … A n …,A N}, and is stored in a matrix SJ, SJ=[{A1, A2, … A n …,A N}]

[0011] Among them, The black-out time length corresponding to the T acceleration classifications of the n-th type of obstacle user, wherein the upper right subscript TM indicates the type of the obstacle user, and the lower right subscript indicates the corresponding acceleration classification.

[0012] Step 2, obtaining the acceleration of the user, calculating the time domain features and frequency domain features of the acceleration;

[0013] Step 3, constructing a probability neural network based on the time domain and frequency domain features of the acceleration, inputting the time domain features and frequency domain features of the acceleration into the probability neural network, and classifying the acceleration.

[0014] Step 4, judging the type of the obstacle user, combining the acceleration classification obtained in step 3, and obtaining the personalized black-out time length of the user through the personalized black-out time length data set.

[0015] Further, the classification method of the obstacle user is as follows:

[0016] Step 1.1, evaluating the hearing, vision and limb flexibility of the user, and the evaluation results of the hearing, vision and limb flexibility include excellent, good, medium, passing and failing;

[0017] Step 1.2, defining the user whose evaluation result of at least one of the hearing, vision and limb flexibility is medium or below as an obstacle user, and dividing the obstacle user into weak hearing medium, passing and failing, weak vision medium, passing and failing, and four-limb flexibility medium, passing and failing according to the degree;

[0018] Step 1.3, considering that the same user may have one or two obstacles of hearing, vision and limb flexibility, the weak hearing, weak vision and four-limb inflexibility with degrees are combined in pairs and in three types, forming 63 types of obstacle users;

[0019] Further, in step 2, the acceleration of the user is obtained, and the time domain features and frequency domain features of the acceleration are calculated, including the following steps:

[0020] Step 2.1, obtaining the acceleration values of the x, y and z axes of the user by using the gravity sensor, and the sampling frequency of the gravity sensor in the application is 200 Hz.

[0021] Step 2.2, calculating the time domain features of the acceleration

[0022] The acceleration values of the x, y and z axes are obtained by using the gravity sensor respectively x a , y a , z a ; then, the acceleration vector area value B in t time is calculated, and the x a , y a , z aThe mean value D, the standard deviation S, and the maximum amplitude and the minimum amplitude M of the acceleration are calculated according to the following formulas (1) to (4).

[0023]

[0024] where t is the sampling time, N1 is the sampling frequency, J x ,J y ,J z are the average acceleration in the three axes at time t, where t = 500 ms and N1 = 200 Hz.

[0025] Step 2.3, select the frequency domain features of acceleration.

[0026] The FFT algorithm is used to extract the frequency domain signals of the virtual reality device x, y, and z axis acceleration. The median frequency E and the average power frequency feature F of the acceleration frequency domain signal are calculated according to the following formulas (5) and (6).

[0027]

[0028] where P(f) is the power spectral density estimation function of acceleration.

[0029] Further, in step 3, the probability neural network is a probability neural network based on Mahalanobis distance, which includes an input layer, a hidden layer, a summation layer, and an output layer.

[0030] Further, in step 1, a segmented data set update strategy is used to construct a personalized black insertion duration data set based on acceleration and obstacle user type;

[0031] The segmented data set update strategy refers to determining the update method of the data set according to the number of users U obtained in each obstacle type data set. The present application sets three stages.

[0032] (1) For the nth obstacle type data set, when 1≤U≤U1, the user performs custom black insertion setting,

[0033]

[0034] where, indicates the black insertion setting of the μth user in the nth obstacle type at the acceleration ;

[0035] In the present application, U1 = 9.

[0036] (2) For the nth obstacle type data set, when U1<U≤U2, the user performs custom black insertion setting

[0037]

[0038] The most suitable black insertion time length matrix obtained by a volunteer test included in the original nth obstacle type data set is

[0039] The black insertion time length data of the nth obstacle type user at the acceleration is:

[0040]

[0041] In the present application, U2=49.

[0042] (3) For the nth obstacle type data set, when U2 The longest black insertion time per second in the U user data is WA, and the shortest black insertion time per second is WB. WA is the starting endpoint of interval C, and WB is the terminal endpoint. Interval C is evenly divided into three subintervals C1, C2, and C3. The U user data is matched with the subintervals. When the user data value is greater than the starting endpoint of the subinterval and less than or equal to the terminal endpoint of the subinterval, the user data is stored in the interval. Calculate the proportion of the number of people in each subinterval to the number of people in interval C. The lowest proportion of the subinterval stores noise data. Remove the user data stored in the subinterval. The user data of the remaining subintervals is retained, and the average value of the data stored in the remaining two intervals is calculated. The average function represents the calculation of the average value. Each acceleration in each data set has two most suitable black insertion time lengths per second, and the user can select according to his own needs. When the number of users who set the black insertion time for each data set reaches U3, the data set stops updating. In the present application, U3=500.

[0043] Beneficial effects: The probability neural network based on the acceleration time domain and frequency domain characteristics can well reflect the differences between different accelerations, reduce the influence of abnormal values on the network accuracy, improve the differences between different characteristics, and thus improve the network accuracy. The Mahalanobis distance is used instead of the Euclidean distance in the probability neural network, which can well calculate the distance between the test sample and the training sample, improve the network recognition accuracy, and does not require normalization processing to improve the network calculation speed.

[0044] Finally, the present application proposes a segmented data set update strategy to construct 63 types of data sets. After determining the user type, the user type and user acceleration are used to realize personalized black insertion display settings for specific users, and the different needs of all types of users are met. BRIEF DESCRIPTION OF DRAWINGS

[0045] Figure 1 Flowchart for user data acquisition;

[0046] Figure 2 This is a diagram of a probabilistic neural network structure based on Mahalanobis distance. Specific Implementation

[0047] This invention is based on existing probabilistic neural networks and artificial intelligence technologies such as time-domain and frequency-domain feature extraction. Traditional acceleration feature extraction methods have the following problems: (1) Directly extracting features from acceleration sequences results in little difference in features between different accelerations; (2) Due to interference from environmental factors, there are many outliers in acceleration sequences, and traditional acceleration feature extraction cannot solve the problem of many outliers; (3) The correlation between various acceleration feature components is not learned, resulting in the inability to improve network recognition accuracy. First, this invention proposes a probabilistic neural network based on acceleration time-domain and frequency-domain features, extracting the time-domain and frequency-domain features of acceleration as the input of the network, reducing the impact of outliers on network accuracy, improving the difference between different features, and thus improving network accuracy. Then, Mahalanobis distance is used instead of Euclidean distance to calculate the correlation between various acceleration feature components; finally, this invention proposes a segmented dataset update strategy to construct a 63-class dataset. After determining the user type, personalized black-out display settings are implemented for specific users based on user type and user acceleration.

[0048] Step 1: Construct a personalized blackboard insertion duration dataset based on acceleration and obstacle user type; where acceleration is divided into T types according to size, and obstacle users have N types;

[0049] The dataset of blackboard insertion duration for different types of users with disabilities is represented as {A1, A2, ... A... n …,A N}, stored in matrix SJ, SJ = [{A1,A2,…A... n …,A N}]

[0050] in, TM represents the black insertion duration for the T acceleration categories corresponding to the nth type of obstacle user, where the superscript of TM indicates the type of obstacle user and the subscript of TM indicates the corresponding acceleration category.

[0051] The specific methods for classifying users with disabilities are as follows:

[0052] Step 1.1: Assess the user's auditory, visual, and physical dexterity.

[0053] The hearing, vision and physical flexibility of the user are respectively evaluated; the evaluation results of the hearing, vision and physical flexibility respectively include excellent, good, medium, passing and failing; when the evaluation score is in the interval [90, 100], it is excellent, in the interval [80, 90), it is good, in the interval [70, 80), it is medium, in the interval [60, 70), it is passing, and less than 60 points is failing.

[0054] Step 1.2, define the user whose evaluation result of at least one of hearing, vision and physical flexibility is medium or below as a disabled user, and the disabled user includes weak hearing medium, passing and failing, weak vision medium, passing and failing, and four-limb flexibility medium, passing and failing according to the degree;

[0055] Step 1.3, considering that the same user may have one or two of hearing, vision and physical flexibility, the weak hearing, weak vision and four-limb inflexibility with degrees are combined in pairs, and the three types are combined to form 63 types of disabled users, as shown in the following table, and N = 63 in the classification method of the application;

[0056] Table 1 User type with only one of hearing, vision and physical flexibility

[0057]

[0058] Table 2 Disabled types obtained by combining weak hearing, weak vision and four-limb inflexibility with degrees in pairs

[0059]

[0060]

[0061] Table 3 Disabled types obtained by combining weak hearing, weak vision and four-limb inflexibility with degrees in three types

[0062]

[0063]

[0064] Step 2, obtain the acceleration of the user, and calculate the time domain features and frequency domain features of the acceleration.

[0065] Step 2.1, obtain the acceleration values of x, y and z axes of the user by using the gravity sensor, and the sampling frequency of the gravity sensor in the application is 200 Hz.

[0066] Step 2.2, calculate the time domain features of the acceleration

[0067] The acceleration values of x, y and z axes are obtained by using the gravity sensor respectively a , ya a Then, the acceleration vector area value B in t time is calculated, and the mean value D, the standard deviation S and the maximum amplitude and the minimum amplitude M of x a a a x y z The average acceleration in t time is J

[0068]

[0069] Wherein, t is the sampling time, N1 is the sampling frequency, J x y z The average acceleration in t time is J

[0070] Step 2.3, select the frequency domain feature of acceleration.

[0071] The FFT algorithm is used to extract the acceleration frequency domain signals of the virtual reality device x, y and z axes. The median frequency E and the average power frequency feature F of the acceleration frequency domain signal are calculated, and the calculation formula is shown in formula (5) and (6).

[0072]

[0073] Wherein, P(f) is the power spectrum density estimation function of acceleration.

[0074] Step 3, construct a probability neural network based on the time domain and frequency domain features of acceleration, input the time domain and frequency domain features of acceleration into the probability neural network, and classify the acceleration. The probability neural network is a probability neural network based on Mahalanobis distance.

[0075] The Euclidean distance is used in the probability neural network to calculate the correlation between the measured data and the sample data, but the data needs to be normalized before each calculation using the Euclidean distance, which is also a defect of the Euclidean distance. The Euclidean distance will be affected by the measurement unit of the data. At the same time, the Euclidean distance cannot calculate the correlation between each feature component. In order to solve this problem, the Mahalanobis distance is used instead of the Euclidean distance to calculate the correlation of the data.

[0076] The probability neural network based on Mahalanobis distance is shown in Figure 2 The O*Q size of the measured data DC is input from the input layer, and the measured data in the application is the time domain and frequency domain features of acceleration. Then, the distance between the measured data and the sample data is calculated through the hidden layer to return a scalar value. In this embodiment, O=3, Q=6, and the Mahalanobis distance is used for calculation. The calculation formula of the Mahalanobis distance is as follows:​​

[0077]

[0078] wherein T = diag (λ1, λ2,..., λ r ), R and F represent unitary matrices, λ l represent unequal singular values, l represents a rank, l e (1, 2,... r), H represents a conjugate transpose, V i represents an input to-be-tested feature matrix, V j represents a sample matrix.

[0079] First, the input of the network is a matrix DC of O*Q size, DC = [dc1, dc2,... dc6] T , wherein T represents a transpose, and the input layer has a total of 6 neurons. They are time domain features and frequency domain features B, D, S, M, E and F of acceleration. The output of the sample layer is EC, EC = [ec 1,1 , ec 1,2 , ec 2,1 , ec 2,2 ..., ec k,j , ec 15,1 , ec 15,2 ], ec k,j represents the jth neuron corresponding to the kth class of the sample layer.

[0080] In this embodiment, k e {1, 2,... 15}, j e {1, 2}.

[0081]

[0082] wherein yb k,j represents a training sample, d is the dimension of the input data, and σ is a hyperparameter.

[0083] In this embodiment, d = 6, σ = 0.1, each feature has 2 training samples, and 15 features have a total of 30 training samples, that is, the sample layer has 30 neurons.

[0084] Then, the summation layer sums the outputs of the neurons of the same class, and the output is

[0085] FC = [fc1,..., fc b ,..., fc 15 ]

[0086] wherein

[0087] In this embodiment, each feature has 2 training samples, that is, ω = 2.

[0088] Finally, the class corresponding to the maximum probability value in the output layer selection sum layer is the class output by the output layer. The calculation formula is as follows:

[0089] Outpurt=Max(FC) (9)

[0090] Step 4, judging the type of the obstacle user, combining the acceleration classification obtained in step 3, obtaining the personalized black insertion time length of the user through the personalized black insertion time length data set.

[0091] Setting the question and answer link, the system evaluates the hearing, vision and limb flexibility of the user, and grades the performance of the above three senses, and divides them into excellent, good, medium, passing and failing in order of good and bad. Finally, the system determines the user type according to the user's answer content.

[0092] In the hearing, vision and limb flexibility evaluation, at least one of them is medium or below, then the personalized black insertion setting is recommended, then the personalized black insertion time length of the user is obtained through the personalized black insertion time length data set combined with the acceleration classification obtained in step 3, but the black insertion data in the data set cannot guarantee to meet the requirements of all users.

[0093] Since each type of acceleration needs to customize the black insertion setting, which requires a lot of time for the user, the user can choose to customize the black insertion setting or skip it. If the user chooses to customize the black insertion setting, the user's customized black insertion data will be stored and can be directly used by logging into the personal account;

[0094] In step 1, based on acceleration, the process of constructing the personalized black insertion time length data set corresponding to different types of obstacle users, since too much data needs to be constructed, the user population that can be found is relatively small, so the user is assisted to construct the data set as shown in Figure 1 Customized black insertion setting.

[0095] (1) If the user customizes the black insertion, the system will store the user's customized black insertion setting in the user library. After the system recognizes the user's motion acceleration, the system automatically matches the black insertion setting in the user library to complete the video black insertion task.

[0096] (2) If the user does not customize the black insertion, the user needs to use the black insertion time length data set constructed by the present application, which includes 63 black insertion time length data sets of different types of obstacle users {A1, A2, …, A 63} each data set includes 15 acceleration corresponding to the length of the black display, by identifying the acceleration of the user with disabilities can be obtained black time. Since the need to establish 63 different levels of different types of data set of users with disabilities, it is difficult to find a number of volunteers to test, which requires a large amount of manpower and material resources. In order to solve this problem, the present application provides a kind of data set updating method, can solve the problem of lack of volunteers when the data set is difficult to establish.

[0097] The present application finds 1 volunteer for testing according to the requirements of each type of data set, a total of 63 volunteers are needed for testing, and it is difficult to find 63 different types of volunteers. When there is no suitable volunteer for the data set, the data set is marked as data unavailable, and the data in the data set is 0. Until the data is stored, the data in the data set can be read. The data set is stored in matrix SJ,

[0098] SJ=[{A1,A2,…,A 63}]

[0099]

[0100] The most suitable black display time per second in the original data set SJ is obtained by one volunteer, which is not universal, and there may be deviation when using the user, resulting in poor user experience. Therefore, the present application uses user-defined black data to update the data set. Before the data set is updated, the amount of user data is not enough, and it is difficult to distinguish useful information and noise information, so the applicability of the data set is not large; when the amount of user data is too much before the data set is updated, it is difficult to update the data set in real time. In order to gradually improve the practicability of the data set, the present application proposes a segmented data set updating strategy.

[0101] Determine whether the user meets the requirements of data set 1 (data set 1 is weak hearing medium), user 1 selects custom black setting, and obtains the custom black data of the user Wherein The upper left mark indicates the first type of user with disabilities, and the lower left mark 2 indicates the second user, The upper left mark and the lower left mark of Indicate the same meaning, The lower right mark of indicates that the acceleration is classified as 1. TJ represents the black data of a user, and the black data of multiple users is stored in TS. TM is the result obtained after the segmented data set updating of TS.

[0102] Take the first type of data set 1 as an example, when U1 users perform custom black setting, Wherein Indicates the acceleration type. In this embodiment, U1∈{1,2,..9}.

[0103] Taking data set 1 as an example, when U2 users perform custom black insertion setting, U2 user black insertion setting data is obtained The most appropriate black insertion duration matrix of volunteer 1 in original data set 1 is obtained through testing In the embodiment, U2∈{10,11,..49}.

[0104] Taking data set 1 as an example, when U3 users perform custom black insertion setting, U3 user black insertion setting data is obtained There are more useful data and noise data in TS. Taking acceleration 1 in data set 1 as an example, the longest black insertion duration per second of U3 user data is WA, and the shortest black insertion duration per second is WB. WA is taken as the starting endpoint of interval C, and WB is taken as the terminal endpoint. Interval C is evenly divided into three subintervals C1, C2 and C3. The data of U3 users is matched with the subintervals. When the size of user data is greater than the starting endpoint of the subinterval and less than or equal to the terminal endpoint of the subinterval, the user data is stored in the interval. The proportion of the number of people in each subinterval to the number of people in interval C is calculated. The noise data is stored in the subinterval with the lowest proportion. The user data stored in the subinterval is removed. The user data in the remaining subintervals is retained, and the average value of the data stored in the remaining two intervals is calculated, The average function represents the calculation of the average value. There are two most appropriate black insertion durations per second for each acceleration in each data set. Users can select according to their own needs. In the embodiment, U3∈{50,11,..500}. When the number of users who perform custom black insertion setting for each data set reaches 500, the data set stops updating.

[0105] The segmented data set updating strategy proposed in the application can effectively utilize the data of each user when the user data is small in the early stage of data set updating and the useful data and noise data cannot be distinguished. The segmented data set updating strategy can divide the user data into useful data and noise data, remove the noise data, and update the data set by using the useful data when the user data increases in the middle and late stages of data set updating.

[0106] ​The application provides a virtual reality black insertion method for personalized data analysis of the handicapped based on acceleration, wherein first, time domain and frequency domain features of acceleration are extracted as inputs of a probability neural network to reduce the influence of abnormal values on the accuracy of the network; then, the Mahalanobis distance is used instead of the Euclidean distance, the Mahalanobis distance does not need to consider the data measurement unit, and meanwhile, the relationship between various feature components can be represented, so that the relationship between the to-be-tested sample and the training sample can be better calculated; finally, a segmented data set updating strategy is provided to construct 63 types of data sets, the user is set with personalized black insertion according to the user type, and the different needs of all types of users are met.

Claims

1. A personalized virtual reality blackboard insertion method for people with disabilities based on acceleration, characterized in that, It includes the following steps: Step 1: Adopt a segmented dataset update strategy to construct a personalized blackout duration dataset based on acceleration and the type of impaired user; where the acceleration is divided into T types according to magnitude, and there are N types of impaired users; The dataset of blackboard insertion duration for different types of users with disabilities is represented as {A1, A2, ... A... n …,A N }, stored in matrix SJ, SJ = [{A1,A2,…A... n …,A N }],in, TM represents the black insertion duration of the T acceleration categories corresponding to the nth type of obstacle user, where the superscript of TM indicates the type of obstacle user and the subscript of TM indicates the corresponding acceleration category; The segmented dataset update strategy refers to determining the update method of the dataset according to the number of users U already obtained in each type of impairment dataset, and three stages are set; (1) For the dataset of the nth type of impairment, when 1 ≤ U ≤ U1, the user makes a custom blackout setting; in, This represents the black insertion setting for the μ-th user at acceleration τ in the n-th obstacle type. (2) For the dataset of the nth type of impairment, when U1 < U ≤ U2, the user makes a custom blackout setting The optimal blackboard insertion duration matrix obtained from the original dataset of the nth obstacle type, including a test conducted by a volunteer, is: Then the blackout duration data of the user of the nth type of impairment at acceleration τ is: (3)For the nth type of obstacle type dataset, when U2 < U ≤ U3, the user performs a custom black insertion setting, and the black insertion setting data of U users are obtained as Among the U user data, the longest black insertion duration per second is WA, and the shortest black insertion duration per second is WB. Taking WA as the starting endpoint of the interval C and WB as the ending endpoint, the interval C is evenly divided into 3 sub-intervals C1, C2, and C3. The U user data are matched with the sub-intervals. When the value of the user data is greater than the starting endpoint of the sub-interval and less than or equal to the ending endpoint of the sub-interval, the user data is stored in that interval; calculate the proportion of the number of people in each sub-interval to the number of people in the interval C. The sub-interval with the lowest proportion stores noise data, and the user data stored in this sub-interval is removed; retain the user data of the remaining sub-intervals, and calculate the average value of the data stored in the remaining two intervals. Among them, the average() function represents calculating the average value. There are two most suitable black insertion durations per second for each acceleration in each dataset, and the user selects according to their own needs. When the number of users with custom black insertion settings in each dataset reaches U3, the dataset stops updating; Step 2: Obtain the acceleration of the user, and calculate the time-domain features and frequency-domain features of the acceleration; Step 3: Construct a probabilistic neural network based on the time-domain and frequency-domain features of acceleration, input the time-domain features and frequency-domain features of acceleration into the probabilistic neural network, and classify the acceleration; Step 4: Determine the type of impaired user, combine the acceleration classification obtained in Step 3, and obtain the personalized blackout duration of the user through the personalized blackout duration dataset.

2. The method for personalized virtual reality black screen insertion for people with disabilities based on acceleration, as described in claim 1, is characterized in that... The classification method of impaired users is as follows: Step 1.1: Evaluate the user's auditory, visual, and limb flexibility. The evaluation results of auditory, visual, and limb flexibility respectively include excellent, good, medium, pass, and fail; Step 1.2: Define users with at least one evaluation result of medium or below in auditory, visual, and limb flexibility as impaired users. According to the degree of impairment, the impaired users are divided into mild hearing impairment medium, pass, and fail, mild vision impairment medium, pass, and fail, and limb flexibility medium, pass, and fail; Step 1.3: Considering that the same user may have one or two impairments in auditory, visual, and limb flexibility, so combine the different degrees of mild hearing impairment, mild vision impairment, and limb inflexibility in pairs, and conduct combinations of three types to form 63 types of impaired users.

3. The method for personalized virtual reality black screen insertion for people with disabilities based on acceleration, as described in claim 1, is characterized in that... In Step 2, the acceleration of the user is obtained, and the time-domain features and frequency-domain features of the acceleration are calculated, [[ID= ​ ​ Acceleration values ​​along the x, y, and z axes were obtained using a gravity sensor. a y a z a Then, calculate the acceleration vector region value B within time t, and calculate x within time t. a y a z a The mean D, standard deviation S, maximum amplitude and minimum amplitude M are calculated using formulas (1) to (4). Where t is the sampling time, N1 is the sampling frequency, and J x J y J z These are the average accelerations on the three axes at time t, where t = 500 ms and N1 = 200 Hz. ​ ​ F=∫0 ∞ fP(f)df / ∫0 ∞ P(f)df (6) ​ 4. The method for personalized virtual reality black screen insertion for people with disabilities based on acceleration, as described in claim 1, is characterized in that... ​

Images