Driving emotion recognition method, system and program product
By collecting driving behavior data through the vehicle's built-in sensors and using a support vector machine model and k-fold cross-validation, a non-invasive, low-cost, and highly accurate method for recognizing driver anger was achieved, solving the problems of privacy infringement and high cost in existing recognition methods.
Patent Information
- Application Number
- CN202510997067.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-18
- Publication Date
- 2025-11-21
AI Technical Summary
Existing technologies for identifying driver anger involve privacy violations, high costs, or poor performance in noisy environments, and lack non-invasive, low-cost, and accurate identification solutions.
Indirect data on driving behavior is collected by the vehicle's built-in sensors, preprocessed and trained using a support vector machine model, and the driver's anger is identified in real time. The model parameters are then optimized using k-fold cross-validation.
It achieves high recognition accuracy while maintaining the system's non-invasiveness and unobtrusiveness, reducing costs and avoiding interference with the driver.
Smart Images
Figure CN120995232A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of intelligent transportation technology, and in particular to driving emotion recognition methods, driving emotion recognition systems and program products. Background Technology
[0002] With the development of intelligent transportation systems, driving safety has become a focus of social concern. A driver's emotional state, especially anger, has a significant impact on driving behavior and may increase the risk of traffic accidents. Current technologies for identifying driver anger mainly include camera-based facial expression analysis, voice emotion recognition, and physiological signal monitoring (such as heart rate and skin conductivity). However, these methods have significant limitations:
[0003] (1) Camera-based facial expression analysis: This requires a high-definition camera to continuously monitor the driver's face, which not only infringes on privacy but may also affect the accuracy of recognition due to issues such as light and obstruction.
[0004] (2) Voice emotion recognition: It relies on the microphone to capture the driver's voice, but it is not effective in noisy environments and is ineffective for drivers who are unwilling to speak or speak in a low voice.
[0005] (3) Physiological signal monitoring: Although it can provide relatively accurate emotional information, it requires wearing a special sensor, which affects driving comfort, and the equipment cost and maintenance cost are high.
[0006] Therefore, existing technologies lack a solution that can effectively identify angry driving emotions without interfering with driving and at the same time protect personal privacy. Summary of the Invention
[0007] The purpose of this invention is to provide a driving emotion recognition method, system, and program product that can accurately identify a driver's anger emotions in a non-invasive and low-cost manner by analyzing indirect data related to driving behavior.
[0008] This invention provides the following solution:
[0009] According to one aspect of the present invention, a driving emotion recognition method is provided, the driving emotion recognition method comprising:
[0010] Using the vehicle's built-in sensors, indirect data on the driver's driving behavior is collected in real time.
[0011] Preprocess indirect data on driving behavior to extract key feature vectors;
[0012] An emotion recognition model is trained using indirect data on driving behavior with known emotion labels.
[0013] New indirect data on driving behavior are input into the trained emotion recognition model in real time, and the model outputs the judgment result of the driver's anger.
[0014] Among them, the indirect data of driving behavior with known emotion labels includes the key feature vectors that have been extracted;
[0015] Emotion recognition models include support vector machine models.
[0016] Optional indirect driving behavior data includes: steering wheel speed, steering wheel pressure, horn frequency, accelerator force, speeding, and braking force.
[0017] Optional key feature vectors include: peak steering speed, peak steering wheel pressure, and number of speeding incidents.
[0018] Optionally, an emotion recognition model can be trained using indirect driving behavior data with known emotion labels, including:
[0019] The model parameters of the emotion recognition model were optimized using k-fold cross-validation.
[0020] The model parameters include: mixing coefficient α, kernel coefficient γ of the Gaussian kernel, degree d of the polynomial kernel, and penalty parameter C.
[0021] Optionally, the model parameters of the emotion recognition model can be fine-tuned using k-fold cross-validation, including:
[0022] For each combination of parameters that needs to be tuned, the optimal solution range for each parameter is determined through a coarse-tuning process.
[0023] Within the optimal solution range of each model parameter, a fine-tuning process is invoked to determine the optimal solution for each parameter.
[0024] Optionally, for each combination of parameters requiring tuning, a coarse-tuning process is used to determine the optimal solution range for each parameter, including:
[0025] The training data is divided into k parts, one of which is used as the validation set;
[0026] After k tests, each time a different validation set was used;
[0027] The results of k models are obtained, and the average value is taken as the final result.
[0028] Optionally, after k tests, a different validation set is used each time, including:
[0029] Obtain different combinations of model parameter values;
[0030] Set the model parameters of the model to be tested to one of the different combinations of values;
[0031] Validate the model using a validation set;
[0032] Obtain the classification accuracy of the model during the validation process;
[0033] Based on whether the classification accuracy meets the requirements, determine whether the current combination of values belongs to the optimal solution interval.
[0034] Optionally, the values of the same parameters in the model parameters form an arithmetic sequence, and the difference between adjacent values in the arithmetic sequence is the set step size.
[0035] According to a second aspect of the present invention, a driving emotion recognition system is provided, the driving emotion recognition system comprising:
[0036] The data acquisition module is used to collect indirect data on the driver's driving behavior in real time using the vehicle's built-in sensors.
[0037] The preprocessing module is used to preprocess indirect data on driving behavior and extract key feature vectors.
[0038] The training module is used to train an emotion recognition model using indirect data of driving behavior with known emotion labels.
[0039] The judgment module is used to input new indirect data on driving behavior into the trained emotion recognition model in real time and output the judgment result of the driver's anger.
[0040] Among them, the indirect data of driving behavior with known emotion labels includes the key feature vectors that have been extracted;
[0041] Emotion recognition models include support vector machine models.
[0042] According to three aspects of the present invention, a computer program product is provided, comprising a computer program that, when executed by a processor, implements the driving emotion recognition method described above.
[0043] The above solution achieves the following beneficial technical effects:
[0044] The implementation of this invention mainly relies on the powerful classification ability of SVM and the rationality of feature selection, which ensures high recognition accuracy while maintaining the non-invasiveness and imperceptibility of the system. Attached Figure Description
[0045] Figure 1 This is a flowchart of a driving emotion recognition method provided in one or more embodiments of the present invention;
[0046] Figure 2 This is a schematic diagram of the principle of a hyperplane provided in one or more embodiments of the present invention;
[0047] Figure 3 This is a flowchart of the training operation in the driving emotion recognition method provided in one or more embodiments of the present invention;
[0048] Figure 4 This is a flowchart of a coarse adjustment operation provided in one or more embodiments of the present invention;
[0049] Figure 5 This is a flowchart of a test in a coarse adjustment operation provided by one or more embodiments of the present invention;
[0050] Figure 6 This is a flowchart of a driving emotion recognition method provided in one or more embodiments of the present invention;
[0051] Figure 7 This is a structural diagram of a driving emotion recognition device provided in one or more embodiments of the present invention. Detailed Implementation
[0052] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0053] Figure 1 This is a flowchart of a driving emotion recognition method provided in one or more embodiments of the present invention. See also... Figure 1 The driving emotion recognition method includes the following steps:
[0054] S11 uses the vehicle's built-in sensors to collect indirect data on the driver's driving behavior in real time.
[0055] S12 preprocesses indirect data on driving behavior and extracts key feature vectors.
[0056] S13 uses indirect data of driving behavior with known emotion labels to train an emotion recognition model.
[0057] S14: Input new indirect data on driving behavior into the trained emotion recognition model in real time, and output the judgment result of the driver's anger.
[0058] Indirect data on driving behavior refers to physical data related to a driver's driving behavior. This data is either generated by the driver's driving behavior or is related to the driver's driving behavior.
[0059] Taking steering wheel pressure as an example, the generation of steering wheel pressure is mainly determined by the driver's driving behavior. Generally speaking, when the driver's attention is focused, the steering wheel pressure generated is greater; when the driver's attention is scattered, the steering wheel pressure generated is less.
[0060] In addition, during driving, the steering wheel pressure is greater when encountering an emergency and less when not encountering an emergency. These emergencies could be pedestrians or vehicles suddenly appearing on the road ahead, vehicles suddenly changing lanes, and so on.
[0061] Taking cab temperature as another example, although cab temperature is not directly generated by driving behavior, statistics show that when the cab temperature is too high, drivers are more prone to irritability. Therefore, cab temperature is data related to the driver's driving behavior.
[0062] It should be noted that, whether generated by or related to the driver's driving behavior, all indirect driving behavior data is raw data directly collected by sensors. This means that the indirect driving behavior data has not undergone any processing or refinement.
[0063] Because indirect driving behavior data is raw, unprocessed data, it inevitably contains some abnormal data that needs to be removed. Due to the presence of this data, a preprocessing operation is required on the raw data, i.e., indirect driving behavior data.
[0064] Besides removing outlier data, preprocessing also includes other operations. For example, for steering wheel pressure data, it's necessary to extract the peak value from the collected, continuous steering wheel pressure waveform. Another example is the need for statistical analysis to extract the number of speeding incidents during driving.
[0065] After the above preprocessing operations, the key feature vectors are obtained. These key feature vectors are the raw data used for training and validating the SVM model, which is a sentiment recognition model. Accurate key feature vectors are required for all training and optimization work on the model.
[0066] After extracting the key feature vectors, the training of the emotion recognition model begins. The most crucial operation in model training is fine-tuning some key parameters to make the model's output, i.e., the classification result, more closely resemble the actual situation.
[0067] It should be noted that in this embodiment, k-fold cross-validation was used during model training.
[0068] The so-called k-fold cross-validation method divides the key feature vectors into k equal parts, uses one part as the validation data for the model, and uses the other k-1 parts as the training data for the model, thus performing training and validation of the model.
[0069] The selected validation data can also be rotated among k key feature vectors, thereby enabling iterative training and tuning of the model.
[0070] It's understandable that if the key feature vector is divided into k parts, the training and validation process described above can be iterated k times. After completing these k iterations, the average of the parameters from the k iterations can be used as the final model parameters. Alternatively, the model parameters from the iteration with the best performance during the k iterations can be used as the final model parameters.
[0071] After k-fold cross-validation, the SVM model training is complete. This SVM model can then be directly applied to the driver's emotion recognition process.
[0072] There are only two possible outcomes for driver emotion recognition: anger or no anger.
[0073] Moreover, in the actual emotion recognition process, the model's input is no longer training data, but actual data collected by the sensors.
[0074] Additionally, it's important to note that unforeseen issues may arise during model usage. This means the model may require real-time tuning during application. Therefore, the applied model should include interfaces for further optimization.
[0075] These interfaces can be function calls used for tuning.
[0076] In this embodiment, the emotion recognition model employs a single kernel function. Furthermore, this single kernel function is a hybrid kernel function combining a Gaussian kernel and a multinomial kernel. The advantage of using a hybrid kernel function is its ability to capture nonlinear relationships at different scales within the driving data distribution.
[0077] Specifically, the hybrid kernel function can be defined as follows:
[0078] K(x i ,x j )=αK RBF (x i ,x j )+(1-α)K POLY (x i ,x j )
[0079] in:
[0080] It is a Gaussian kernel function.
[0081] It is a polynomial kernel function.
[0082] α is the mixing coefficient, used to balance the contributions of the Gaussian kernel and the polynomial kernel.
[0083] γ is the kernel coefficient of the Gaussian kernel.
[0084] d is the degree of the polynomial kernel.
[0085] Figure 2 The principle of the SVM model provided in this embodiment is illustrated. In this embodiment, the SVM model is essentially a classifier capable of distinguishing between situations that belong to the emotion of anger and situations that do not.
[0086] Figure 3 This is a flowchart of the training operation in the driving emotion recognition method provided in one or more embodiments of the present invention. See also Figure 3 The model parameters of the emotion recognition model were tuned using k-fold cross-validation, including:
[0087] S31. For each combination of parameters that needs to be tuned, the optimal solution range for each parameter is determined through a coarse-tuning process.
[0088] S32, within the optimal solution range of each model parameter, calls the fine-tuning process to determine the optimal solution for each parameter.
[0089] As described in the foregoing embodiments of this invention, a significant portion of the model training process in this application involves tuning the model parameters. This tuning phase can be further subdivided into coarse-tuning and fine-tuning processes.
[0090] The purpose of the coarse-tuning process is to determine the optimal solution range for each parameter; while the purpose of the fine-tuning process is to further determine the optimal solution from the optimal solution range for each parameter.
[0091] It's understandable that the optimal solution interval is a range of values within the numerical solution space. Within this range, there may be an infinite number of solutions. However, whether these solutions represent the final optimal solution needs to be determined through fine-tuning the parameters.
[0092] The reason for adopting the optimization strategy of coarse tuning first and then fine tuning is that we hope to accurately find the optimal values of the model parameters, avoid spending too much effort on fine tuning in invalid value ranges, and improve the efficiency of model optimization.
[0093] Whether it's coarse or fine adjustment, the adjustment process for each parameter is completed based on the set adjustment step size.
[0094] Without losing generality, let's take steering wheel pressure adjustment as an example. In the coarse adjustment stage, the adjustment step can be set to 0.5 Newtons; while in the fine adjustment stage, the adjustment step can be set to 0.1 Newtons.
[0095] Typically, during the coarse-tuning stage, the parameter adjustment step size is set to a relatively large granularity. During the fine-tuning stage, the parameter adjustment step size is set to a relatively small granularity.
[0096] In this embodiment, whether in the coarse-tuning stage or the fine-tuning stage, the parameters that need to be adjusted in the SVM model are: the mixing coefficient α, the kernel coefficient γ of the Gaussian kernel, the degree d of the polynomial kernel, and the penalty coefficient C of the SVM.
[0097] Typical search step sizes for each model parameter are as follows:
[0098] Initially define the search range for parameters α, γ, d, and C:
[0099] α∈[0, 1] (step size is 0.1);
[0100] γ∈[10 -3 10 3 (Uniformly distributed on the logarithmic scale, with a value of 10) -3 10 -2 ,…,10 2 10 3 );
[0101] C∈[1,100] (where 1,10,20,…,90,100);
[0102] d∈[1,5] (step size is 1).
[0103] Figure 4 This is a flowchart of a coarse adjustment operation provided in one or more embodiments of the present invention. See also... Figure 4 For each parameter combination that needs optimization, a coarse-tuning process is used to determine the optimal solution range for each parameter, including:
[0104] S41, divide the training data into k parts, one of which is used as the validation set.
[0105] S42, after k tests, each time with a different validation set.
[0106] S43, obtain the results of k sets of models, and take the average value as the final result.
[0107] This embodiment uses the coarse-tuning process as an example to illustrate the application of the k-fold cross-validation method in the process of model parameter tuning.
[0108] The core of k-fold cross-validation lies in using the same dataset to complete both model training and validation.
[0109] First, the original training data is divided into k parts. This division can be an equal division in terms of data volume. That is, after the division, the k parts of training data are equal in size.
[0110] After completing the k-part partitioning, one part of the k-part data is used as the validation set, and the other k-1 parts are used as the training set.
[0111] After selecting the training and validation sets, you can perform model testing, which is the training and tuning of the SVM model.
[0112] After completing one model test, a new validation set is selected from the k training data sets. After selecting the new validation set, the remaining k-1 data sets are used as the training set. The test is then performed again.
[0113] This process is repeated until k tests are completed. The interval containing the mean of the model parameters obtained in each test is taken as the optimal solution interval for the model parameters.
[0114] For example, after all the tests, the mean of α is 3.04. Therefore, the interval containing 3.04, that is, between 3 and 4, can be taken as the optimal solution interval for α.
[0115] It should be noted that this embodiment only illustrates the application process of k-fold cross-validation in the coarse-tuning of model parameters. The application process of k-fold cross-validation in the fine-tuning of the model is similar to that provided in this embodiment, and will not be repeated here.
[0116] Figure 5 This is a flowchart of a single test in the coarse-tuning operation provided by one or more embodiments of the present invention. See also... Figure 5 After k tests, each time with a different validation set, the following steps were performed:
[0117] S51, obtain different combinations of model parameter values.
[0118] S52 sets the model parameters of the model to be tested to one of the different combinations of values.
[0119] S53 uses a validation set to validate the model.
[0120] S54, obtain the classification accuracy of the model during the validation process.
[0121] S55. Based on whether the classification accuracy meets the requirements, determine whether the current combination of values belongs to the optimal solution interval.
[0122] This embodiment uses a test during the coarse-tuning of the model's parameters as an example to illustrate the specific process of each test.
[0123] Because an adjustment step size is set for each model parameter that needs adjustment before coarse tuning, the possible values of each model parameter that needs adjustment are finite during the coarse tuning process. Moreover, the total number of model parameters that need adjustment is also finite, so the number of combinations of different values of the model parameters is also finite.
[0124] Taking the parameters that need to be adjusted in this application embodiment as an example, there are a total of 4 parameters that need to be adjusted, namely the mixing coefficient α, the kernel coefficient γ of the Gaussian kernel, the degree d of the polynomial kernel, and the penalty coefficient C of the SVM.
[0125] The mixing coefficients can take 11 possible values from 0 to 1. The kernel function γ of the Gaussian kernel can take 10 values. -3 Up to 10 3 There are a total of 7 point values. The degree d of the polynomial can take 11 point values from 0 to 100. The penalty coefficient has 6 point values from 1 to 5.
[0126] Combining these possible point values constitutes the set of possible values for the model parameters during the coarse-tuning process.
[0127] In a single test, the model's input data is fixed, meaning the validation set has already been selected. However, the model's parameters are not fixed; they can be any value from the set of possible values.
[0128] In a single test, it is necessary to validate every possible value in the set of possible values for the model parameters. That is, the model parameters need to be adjusted to each possible value, and then, for each of these values, the model is input with validation data from the validation set to test the model's classification accuracy.
[0129] After validating the model in each possible set of values, the combination of values with the best classification accuracy is selected to determine the optimal solution range for the model parameters.
[0130] If the model has pre-defined convergence conditions, we can check whether the model's classification accuracy meets the pre-defined model convergence conditions, that is, whether the actual classification accuracy reaches the pre-defined accuracy threshold, in order to determine whether the combination of values in this test belongs to the optimal solution range that needs to be achieved.
[0131] It should be noted that this embodiment uses a test process during the coarse adjustment process as an example to illustrate the testing process. The test process during the fine adjustment process is similar to the test process described in this embodiment, and will not be repeated here.
[0132] Figure 6 This is a flowchart of a driving emotion recognition method provided in one or more embodiments of the present invention. See also... Figure 6 The driving emotion recognition method includes the following steps:
[0133] S601, Obtain angry driving classification data samples.
[0134] S602, obtain test samples from the acquired angry driving classification data samples.
[0135] S603, obtain training samples from the acquired angry driving classification data samples.
[0136] S604 sets the initial parameters of the SVM model.
[0137] S605 adjusts the parameters of the SVM model based on a fixed step size.
[0138] S606 uses training samples to train the parameter-adjusted SVM model.
[0139] S607, obtain the predicted values of the training samples.
[0140] S608, determine whether the convergence condition is met.
[0141] S609. If the convergence condition is met, determine the optimized values of the parameters.
[0142] S610. If the convergence condition is not met, update the parameters of the SVM model and perform training again.
[0143] S611. Based on the determined parameter optimization values, determine the SVM model for classifying angry driving, and validate the SVM model using test samples.
[0144] In this embodiment, angry driving classification data samples are first obtained. Then, according to the data partitioning principle set in the k-fold cross-validation method, test samples and training samples are obtained from the angry driving classification data samples respectively.
[0145] Next, the SVM model will be trained using the acquired training samples. During training, the initial parameters of the SVM model will be set first, which is to initialize the SVM model.
[0146] Then, the SVM model parameters are adjusted according to a fixed step size. Next, the SVM model is trained using the acquired training data. The training process is essentially a process of further optimizing the model parameters.
[0147] After training, the model is evaluated based on whether the predicted sample values meet the convergence criteria to determine if further training is needed. If the convergence criteria are met, the current parameter values are determined to be the final optimized values for the model. If the convergence criteria are not met, the model parameters need to be adjusted, and further model training needs to be performed.
[0148] Finally, the model tuning results were further validated using the previously determined test dataset.
[0149] Figure 7 A structural diagram of the driving emotion recognition system provided in an embodiment of this application is shown. See also... Figure 7 The driving emotion recognition system includes:
[0150] The data acquisition module 71 is used to collect indirect data on the driver's driving behavior in real time using the vehicle's built-in sensors.
[0151] The preprocessing module 72 is used to preprocess the indirect data of driving behavior and extract key feature vectors.
[0152] Training module 73 is used to train an emotion recognition model using indirect data of driving behavior with known emotion labels.
[0153] The judgment module 74 is used to input new indirect data on driving behavior into the trained emotion recognition model in real time and output the judgment result of the driver's anger.
[0154] Among them, the indirect data of driving behavior with known emotion labels includes the key feature vectors that have been extracted;
[0155] Emotion recognition models include support vector machine models.
[0156] Optional indirect driving behavior data includes: steering wheel speed, steering wheel pressure, horn frequency, accelerator force, speeding, and braking force.
[0157] Optional key feature vectors include: steering wheel speed, peak steering wheel pressure, horn frequency, accelerator pedal pressure, number of speeding incidents, and braking force.
[0158] Optionally, training module 73 is specifically used for:
[0159] The model parameters of the emotion recognition model were optimized using k-fold cross-validation.
[0160] The model parameters include: mixing coefficient α, kernel coefficient γ of the Gaussian kernel, degree d of the polynomial kernel, and penalty parameter C.
[0161] Optionally, the model parameters of the emotion recognition model can be fine-tuned using k-fold cross-validation, including:
[0162] For each combination of parameters that needs to be tuned, the optimal solution range for each parameter is determined through a coarse-tuning process.
[0163] Within the optimal solution range of each model parameter, a fine-tuning process is invoked to determine the optimal solution for each parameter.
[0164] Optionally, for each combination of parameters requiring tuning, a coarse-tuning process is used to determine the optimal solution range for each parameter, including:
[0165] The training data is divided into k parts, one of which is used as the validation set;
[0166] After k tests, each time a different validation set was used;
[0167] The results of k models are obtained, and the average value is taken as the final result.
[0168] Optionally, after k tests, a different validation set is used each time, including:
[0169] Obtain different combinations of model parameter values;
[0170] Set the model parameters of the model to be tested to one of the different combinations of values;
[0171] Validate the model using a validation set;
[0172] Obtain the classification accuracy of the model during the validation process;
[0173] Based on whether the classification accuracy meets the requirements, determine whether the current combination of values belongs to the optimal solution interval.
[0174] Optionally, the difference between different combinations of model parameter values, and between identical parameters, is set as the step size.
[0175] It is worth noting that although only some basic functional modules are disclosed in the embodiments of this invention, it does not mean that the composition of this system is limited to the above-mentioned basic functional modules. On the contrary, what this embodiment intends to express is that, based on the above-mentioned basic functional modules, those skilled in the art can arbitrarily add one or more functional modules in combination with existing technology to form an infinite number of embodiments or technical solutions. That is to say, this system is open rather than closed. The fact that this embodiment only discloses a few basic functional modules should not be considered as the scope of protection of the claims of this invention being limited to the disclosed basic functional modules. At the same time, for the convenience of description, the above device is described separately according to its functions as various units and modules. Of course, in implementing this invention, the functions of each unit and module can be implemented in one or more software and / or hardware.
[0176] The present invention also provides a computer-readable storage medium storing a computer program executable by a vehicle, which, when run on the vehicle, causes the vehicle to perform the steps of a driving emotion recognition method.
[0177] Specifically, the computer storage medium in this embodiment of the invention can be any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. For example, a computer-readable storage medium can be—but is not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this embodiment, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.
[0178] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.
Claims
1. A driving emotion recognition method, characterized in that, The driving emotion recognition method includes: Using the vehicle's built-in sensors, indirect data on the driver's driving behavior is collected in real time. Preprocess indirect data on driving behavior to extract key feature vectors; An emotion recognition model is trained using indirect data on driving behavior with known emotion labels. New indirect data on driving behavior are input into the trained emotion recognition model in real time, and the model outputs the judgment result of the driver's anger. Among them, the indirect data of driving behavior with known emotion labels includes the key feature vectors that have been extracted; Emotion recognition models include support vector machine models.
2. The method according to claim 1, characterized in that, Indirect data on driving behavior includes: steering wheel speed, steering wheel pressure, horn frequency, accelerator force, speeding, and braking force.
3. The method according to claim 1, characterized in that, Key feature vectors include: peak steering speed, peak steering wheel pressure, and number of speeding incidents.
4. The method according to claim 1, characterized in that, Using indirect data on driving behavior with known emotion labels, an emotion recognition model is trained, including: The model parameters of the emotion recognition model were optimized using k-fold cross-validation. The model parameters include: mixing coefficient α, kernel coefficient γ of the Gaussian kernel, degree d of the polynomial kernel, and penalty parameter C.
5. The method according to claim 4, characterized in that, The model parameters of the emotion recognition model are tuned using k-fold cross-validation, including: For each combination of parameters that needs to be tuned, the optimal solution range for each parameter is determined through a coarse-tuning process. Within the optimal solution range of each model parameter, a fine-tuning process is invoked to determine the optimal solution for each parameter.
6. The method according to claim 5, characterized in that, For each parameter combination requiring optimization, a coarse-tuning process is used to determine the optimal solution range for each parameter, including: The training data is divided into k parts, one of which is used as the validation set; After k tests, each time a different validation set was used; The results of k models are obtained, and the average value is taken as the final result.
7. The method according to claim 6, characterized in that, After k tests, each time a different validation set was used, including: Obtain different combinations of model parameter values; Set the model parameters of the model to be tested to one of the different combinations of values; Validate the model using a validation set; Obtain the classification accuracy of the model during the validation process; Based on whether the classification accuracy meets the requirements, determine whether the current combination of values belongs to the optimal solution interval.
8. The method according to claim 7, characterized in that, The values of the same parameters in the model parameters form an arithmetic sequence, and the difference between adjacent values in the arithmetic sequence is the set step size.
9. A driving emotion recognition system, characterized in that, The driving emotion recognition system includes: The data acquisition module is used to collect indirect data on the driver's driving behavior in real time using the vehicle's built-in sensors. The preprocessing module is used to preprocess indirect data on driving behavior and extract key feature vectors. The training module is used to train an emotion recognition model using indirect data of driving behavior with known emotion labels. The judgment module is used to input new indirect data on driving behavior into the trained emotion recognition model in real time and output the judgment result of the driver's anger. Among them, the indirect data of driving behavior with known emotion labels includes the key feature vectors that have been extracted; Emotion recognition models include support vector machine models.
10. A computer program product, characterized in that, It includes a computer program that, when executed by a processor, implements the driving emotion recognition method according to any one of claims 1 to 8.