A vehicle turn signal self-activation method and system

Abstract

The application relates to the technical field of intelligent auxiliary driving, and provides a vehicle turn signal self-starting method and system.A fusion algorithm, such as a Bayesian network and a Kalman filter, is built in a steering decision model based on a deep learning algorithm.When line-of-sight features, steering wheel features and vehicle speed features are extracted, the reliability and correlation of different data are considered, the probability relationship between various factors is considered, the weights of the factors in a final result are dynamically adjusted, different steering intentions of a user are represented and predicted through a probability model, more reliable prediction results are provided, the accuracy and reliability of steering intention recognition are improved, the turn signal can be automatically turned on according to the line-of-sight direction intention of the driver through the cooperative work of a driver monitoring system and rotation monitoring (vehicle speed data and steering wheel data), the convenience and automation level of operation are improved, traffic accidents caused by improper operation of the turn signal are reduced, and road traffic safety is ensured.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent auxiliary driving, and in particular to a vehicle turn signal self-starting method and system. BACKGROUND

[0002] In the modern traffic environment, the vehicle turn signal is a vital safety device. Its main function is to convey the intention of the vehicle to change the driving direction to other road users (including vehicle drivers, pedestrians, etc.), such as vehicle turning, lane changing or parking, etc. When the turn signal is turned on in advance, the surrounding traffic participants have enough time to respond, thereby avoiding traffic accidents.

[0003] However, the current operation of the vehicle turn signal mainly depends on the driver to manually turn it on, which has many defects. On the one hand, manually operating the turn signal requires the driver to additionally distract the turn signal control device while driving (such as turning or lane changing operation), which affects the driving efficiency to some extent. For example, in complex traffic conditions (such as busy intersections or emergency lane changes when driving fast), the driver may forget or delay turning on the turn signal because of focusing on the road conditions, which may mislead other road users and increase the risk of traffic accidents. On the other hand, since the operation of the turn signal is completely dependent on the subjective consciousness of the driver, some drivers may not use the turn signal due to negligence or bad driving habits, which also poses a serious threat to traffic safety.

[0004] In summary, the existing manual operation mode of the vehicle turn signal is low in efficiency and easy to be ignored by the driver, and there are traffic safety problems caused by improper operation of the turn signal. SUMMARY

[0005] The present application provides a vehicle turn signal self-starting method and system, which solves the technical problem of low efficiency and easy to be ignored by the driver in the existing manual operation mode of the vehicle turn signal, and the traffic safety hazard caused by improper operation of the turn signal.

[0006] To solve the above technical problems, the present application provides a vehicle turn signal self-starting method, comprising:

[0007] acquiring user state information in real time based on a driver monitoring system;

[0008] collecting vehicle state information of the current vehicle in real time;

[0009] constructing a turning decision model based on a deep learning algorithm, inputting the user state information and the vehicle state information into the turning decision model to perform feature fusion processing and classification prediction, so as to obtain the turning intention of the user;

[0010] The turning light self-starting control is determined to be executed according to the turning intention of the user.

[0011] In further embodiments, the user state information is acquired in real time based on a driver monitoring system, including:

[0012] The surrounding space of the vehicle is divided into regions, and target regions corresponding to different turning intentions are obtained;

[0013] The head image of the user is collected in real time based on a driver monitoring system, and image recognition is performed to realize gaze tracking, and the gaze tracking is matched with the target regions to determine the user gaze information;

[0014] The user gaze information includes one or more of a gaze angle, a gaze dwell time, a gaze movement speed, and a gaze movement direction.

[0015] The scheme constructs a driver gaze tracking process based on the gaze interaction function of the DMS system, thereby accurately tracking the eye movement of the driver during vehicle driving, and through analysis of the fixation point and fixation time of the user, the gaze attention direction of the driver is recognized in real time, the automatic detection and recognition of the automatic starting of the turning light are realized without contact, the hands are completely liberated, and the driving safety is improved; meanwhile, the surrounding space of the vehicle is divided into regions for gaze tracking, and target regions corresponding to different turning intentions are obtained, so that the user intention (such as whether to look at the left or right rearview mirror, and the left front or right front area, to obtain the possible turning or lane changing intention related information of the driver) is quickly determined through the matching of the gaze and the target region, and the sensitivity of the device is further improved.

[0016] In further embodiments, it further includes: performing pupil positioning and corneal reflection point detection using an infrared sensor to assist image recognition to realize gaze tracking; and / or, obtaining the head posture of the user and estimating the three-axis posture of the head of the user using a deep learning algorithm, and then dynamically compensating the gaze tracking realized by image recognition in combination with the three-axis posture.

[0017] When performing gaze tracking, the scheme synchronously integrates other auxiliary means, including but not limited to performing pupil positioning and corneal reflection point detection using an infrared sensor to assist image recognition to realize gaze tracking; and / or, obtaining the head posture of the user and estimating the three-axis posture of the head of the user using a deep learning algorithm, and then dynamically compensating the gaze tracking realized by image recognition in combination with the three-axis posture, thereby improving the accuracy and reliability of the gaze tracking module.

[0018] In further embodiments, the vehicle state information of the current vehicle is collected in real time, specifically: the vehicle body domain control system is used to continuously monitor the vehicle and obtain corresponding vehicle body information, and the vehicle speed data and steering wheel data are obtained from the vehicle body information as the vehicle state information of the current vehicle.

[0019] The scheme continuously monitors the vehicle by using the vehicle body domain control system and obtains vehicle body information such as vehicle speed data and steering wheel data, further identifies the user's steering intention from the vehicle operation based on the changing parameters of the equipment on the vehicle, and combines the user state information and the vehicle state information for mutual verification, multi-factor fusion intention recognition, and higher recognition accuracy.

[0020] In further embodiments, the steering decision model comprises an input layer, a convolutional layer, a pooling layer, a fully connected layer, and an activation function.

[0021] The input layer is used to normalize the user state information and the vehicle state information to obtain corresponding input features.

[0022] The convolutional layer is used to perform convolutional operations on the input features through a plurality of convolutional kernels to extract features and obtain corresponding gaze features, steering wheel features, and vehicle speed features.

[0023] The pooling layer is used to downsample the gaze features, steering wheel features, and vehicle speed features, reduce the feature dimension, and output corresponding feature sequences.

[0024] The fully connected layer has a built-in fusion algorithm for performing feature fusion processing and classification prediction based on the gaze features, steering wheel features, and vehicle speed features to obtain the user's steering intention.

[0025] The scheme uses a deep learning algorithm to establish a steering decision model, which can adapt to different users through feature extraction, feature fusion, and classification prediction, and has strong pertinence and good adaptability. The deep learning algorithm can learn the user's preferences and behavior patterns to more accurately predict the user's next operation, thereby ensuring the accuracy of the prediction of the user's steering intention and providing good turn signal self-starting services.

[0026] In further embodiments, when the fusion algorithm is a Bayesian network, the feature fusion processing and classification prediction based on the gaze features, steering wheel features, and vehicle speed features to obtain the user's steering intention comprises:

[0027] Classifying scenes based on different driving scenes to set corresponding steering intentions.

[0028] Taking the gaze features, steering wheel features, and vehicle speed features obtained based on the user state information and the vehicle state information as observation variables, and combining the Bayesian theorem to calculate the prior probability corresponding to each steering intention.

[0029] Comparing the prior probabilities of all the steering intentions, and taking the steering intention with the highest probability value as the classification prediction result and outputting it.

[0030] The scheme embeds a fusion algorithm in the steering decision model based on a deep learning algorithm, such as a Bayesian network and a Kalman filter. When the line-of-sight feature, the steering wheel feature, and the vehicle speed feature are extracted, the reliability and correlation of different data are considered, the probability relationship between various factors is considered, the weights of the factors in the final result are dynamically adjusted, the different steering intentions of the user are represented and predicted through a probability model, and more reliable prediction results are provided to improve the accuracy and reliability of steering intention recognition.

[0031] In further embodiments, whether to perform steering lamp self-starting control is determined according to the steering intention of the user, including:

[0032] The steering intention of the user and the corresponding prior probability are obtained from the steering decision model, it is determined whether the prior probability is greater than a preset threshold, if yes, it is determined to perform steering lamp self-starting control, and then a corresponding control instruction is generated based on the steering intention to drive the corresponding left / right steering lamp to start.

[0033] The scheme sets a probability threshold as a preset threshold based on an actual application environment, determines whether to perform steering lamp self-starting control by comparing whether the prior probability of the output steering intention is greater than the preset threshold, and thus the probability of false triggering of the steering lamp can be reduced and driving safety can be improved.

[0034] In further embodiments, the steering decision model is further trained, and the model training includes:

[0035] Driving data is collected and labeled, and the data labeling includes line-of-sight features, steering wheel features, vehicle speed features, and corresponding steering intention labels.

[0036] The driving data is divided into a training set, a validation set, and a test set according to a preset ratio.

[0037] The training set data is used to train the steering decision model, and the model parameters θ are optimized and the loss function is minimized through a back propagation algorithm.

[0038] The test set is used to evaluate the trained steering decision model, and the evaluation criteria of the model evaluation include one or more indicators of accuracy, recall rate, and F1 value.

[0039] The scheme collects a large amount of labeled driving data for training, and the availability of big data enables the model to learn the underlying patterns and reduce overfitting, and thus more complex input data and task scenarios can be processed to achieve higher prediction accuracy and classification effect.

[0040] The application further provides a vehicle turn signal self-starting system, comprising a memory, a processor and a computer program stored on the memory and executable on the processor, and when the processor executes the computer program, a vehicle turn signal self-starting method as described above is implemented.

[0041] In further embodiments, a data acquisition module connected to the processor is further included, and the data acquisition module comprises an image acquisition module, a vehicle speed acquisition module and a steering angle acquisition unit.

[0042] The image acquisition module is configured to acquire user state information in real time.

[0043] The vehicle speed acquisition unit is configured to acquire vehicle speed data.

[0044] The steering angle acquisition unit is configured to acquire steering wheel data of the vehicle.

[0045] The present application is based on the driver monitoring system and body domain control system of the vehicle itself, and defines the image acquisition module, vehicle speed acquisition module and steering angle acquisition unit to acquire data, without the need to install additional equipment, and is applicable to different types of vehicles and has good applicability.

[0046] The beneficial effects of the present application are as follows:

[0047] 1. Improve the automation level of turn signal operation

[0048] Through the cooperative work of the driver monitoring system (DMS) and the rotation monitoring (vehicle speed data and steering wheel data), the turn signal can be automatically turned on according to the driver's line of sight direction intention. This greatly reduces the steps of manual operation of the driver, improves the convenience and automation level of operation.

[0049] 2. Enhance traffic safety

[0050] Due to the reduction of manual operation of the turn signal, the situation of not turning on the turn signal due to the driver's forgetfulness or untimely operation is avoided. When the vehicle turns or changes lanes, turning on the turn signal correctly is crucial for conveying the driving intention to other road users. The present technology can effectively reduce traffic accidents caused by improper operation of the turn signal and ensure road traffic safety.

[0051] 3. Improve the comfort of driving experience

[0052] The driver does not need to be distracted to manually operate the turn signal, especially in complex road conditions that require frequent turning or lane changing, and can focus more on driving operations themselves, such as controlling the vehicle speed and maintaining the vehicle distance. By realizing the automatic operation of the turn signal to reduce the operation burden of the driver, the comfort of the driving experience is improved.

[0053] 4. Improve the accuracy of turning intention recognition

[0054] Various data fusion algorithms such as Kalman filter, Bayesian network and other fusion algorithms are adopted, and then the weights can be dynamically adjusted according to the reliability and correlation of different data. In a complex scene, a classifier (such as a support vector machine, a decision tree or a neural network) can be trained by using a machine learning or deep learning algorithm to distinguish different steering intentions. The recognition accuracy of the driver's steering intention is effectively improved, the possibility of misjudgment is reduced, and the reliability of the automatic turn signal function is further improved. BRIEF DESCRIPTION OF DRAWINGS

[0055] Figure 1 is a vehicle turn signal self-starting method system flowchart provided by the embodiment of the present application;

[0056] Figure 2 is an algorithm execution schematic diagram provided by the embodiment of the present application;

[0057] Figure 3 is a line-of-sight range area diagram provided by the embodiment of the present application;

[0058] Figure 4 is a line-of-sight collection schematic diagram provided by the embodiment of the present application. DETAILED DESCRIPTION

[0059] The embodiments of the present application will be described in detail below with reference to the accompanying drawings. The embodiments are given only for illustrative purposes, and cannot be understood as limiting the present application. The accompanying drawings are only for reference and illustration, and do not constitute a limitation on the scope of protection of the present application, because many changes can be made to the present application without departing from the spirit and scope of the present application.

[0060] The vehicle turn signal self-starting method provided by the embodiment of the present application is as shown in Figures 1-4 In the embodiment, it includes:

[0061] S1, acquiring user state information based on a driver monitoring system in real time; real-time collection of vehicle state information of the current vehicle;

[0062] In the embodiment, the user state information is acquired based on the driver monitoring system in real time, which includes:

[0063] a1, dividing the space around the vehicle into regions to obtain target regions corresponding to different steering intentions respectively;

[0064] Referring to Figure 3 , the space around the vehicle based on the driver's line-of-sight range is divided into regions to obtain target regions including a front line-of-sight region, a left front line-of-sight region, a right front line-of-sight region, a left rearview mirror line-of-sight region and a right rearview mirror line-of-sight region.

[0065] a2, based on the driver monitoring system, collecting the head image of the user in real time, performing image recognition to realize line-of-sight tracking, and performing regional matching with the target region to determine the line-of-sight information of the user;

[0066] The line-of-sight information of the user includes one or more of a line-of-sight angle, a line-of-sight dwell time, a line-of-sight movement speed, and a line-of-sight movement direction.

[0067] Referring to Figure 4 The line-of-sight collection schematic diagram, the line-of-sight interaction function based on the DMS system constructs the driver line-of-sight tracking mechanism, and then during vehicle driving, the eye movement of the driver is accurately tracked and the fixation point is detected, and the line-of-sight attention direction of the driver is identified in real time, such as whether to look at the left or right side mirror, and the left or right front side area.

[0068] The line-of-sight tracking process of the driver is constructed based on the line-of-sight interaction function of the DMS system, so that during vehicle driving, the eye movement of the driver is accurately tracked, the line-of-sight attention direction of the driver is identified in real time by analyzing the fixation point and fixation time of the user, and the automatic detection and identification of the automatic starting of the turn signal are realized without contact, so as to completely liberate the hands and improve the driving safety; at the same time, the line-of-sight tracking is performed to divide the space around the vehicle into regions, and target regions corresponding to different turning intentions are obtained, so that the user intention (such as whether to look at the left or right side mirror, and the left or right front side area, to obtain the possible turning or lane changing intention related information of the driver) is quickly determined through the matching of the line-of-sight and the target region, and the sensitivity of the device is further improved.

[0069] In this embodiment, the collection of the user state information also includes:

[0070] The data of other sensors are fused, such as the pupil positioning and corneal reflection point detection performed by the infrared sensor under low light conditions, to assist the image recognition to realize the line-of-sight tracking; wherein, the pupil positioning, corneal reflection point detection, etc. are performed at the same time in combination with multi-task learning, which can improve the accuracy of line-of-sight estimation.

[0071] And / or, in the case of free movement of the driver's head, the system can dynamically compensate for the line-of-sight estimation, obtain the head posture of the user and estimate the three-axis posture (Yaw, Pitch, Roll) of the user's head using a deep learning algorithm, and then dynamically compensate for the line-of-sight tracking realized by the image recognition in combination with the three-axis posture (i.e. adjust the estimation of the line-of-sight direction according to these posture information).

[0072] The scheme synchronously integrates other auxiliary means when performing gaze tracking, including but not limited to using an infrared sensor to perform pupil positioning and corneal reflection point detection to assist image recognition to realize gaze tracking; and / or, acquiring a user head posture and using a deep learning algorithm to estimate a three-axis posture of the user head, and then combining the three-axis posture to dynamically compensate for gaze tracking realized by image recognition, thereby improving the accuracy and reliability of the gaze tracking module.

[0073] In the embodiment, the vehicle state information of the current vehicle is collected in real time, specifically: the vehicle body domain control system continuously monitors the vehicle and acquires corresponding vehicle body information, and the speed data and steering wheel data are acquired from the vehicle body information as the vehicle state information of the current vehicle.

[0074] The vehicle body domain control system is responsible for continuously monitoring the rotation of the vehicle during driving, accurately detecting the steering angle of the steering wheel, the vehicle speed and other data, and providing necessary data support for judging the steering intention of the driver.

[0075] The scheme further identifies the steering intention of the user from the vehicle operation based on the changing parameters of the device on the vehicle, and verifies each other by combining the user state information and the vehicle state information, so that the multi-factor fusion intention recognition has higher recognition accuracy.

[0076] S2, constructing a steering decision model based on a deep learning algorithm, inputting the user state information and the vehicle state information into the steering decision model to perform feature fusion processing and classification prediction, to obtain the steering intention of the user;

[0077] In the embodiment, the steering decision model includes an input layer, a convolution layer, a pooling layer, a full connection layer and an activation function.

[0078] The input layer is used for normalizing the user state information and the vehicle state information to obtain corresponding input features; the normalization processing is used to map feature values of different dimensions and different ranges to the same scale interval, such as [0, 1], so that the subsequent algorithm can be processed more effectively.

[0079] For example, the feature sequence dimension of the normalized user state information and the vehicle state information is (T, F), where T is the time series length and F is the feature dimension.

[0080] The convolution layer is used to perform convolution operation on the input features through a plurality of convolution kernels to realize feature extraction and obtain corresponding gaze features, steering wheel features and speed features.

[0081] Wherein, the convolution kernel size is (k, F), the step is 1, and the padding mode is the same, wherein the output feature sequence length is kept the same as the input. The convolution layer output feature dimension is (T, C), wherein C is the number of convolution kernels, and k is the width of the convolution kernel on the time sequence, which determines the range covered by the convolution kernel on the time sequence.

[0082] The pooling layer is used for downsampling the line-of-sight feature, steering wheel feature and vehicle speed feature, reducing the feature dimension, and outputting the corresponding feature sequence.

[0083] Wherein, the pooling layer adopts maximum pooling, and the pooling window size is 2 and the step is 2. The pooling layer output feature dimension is (T / 2, C).

[0084] The full connection layer is built-in fusion algorithm, which is used for performing feature fusion processing and classification prediction according to the line-of-sight feature, steering wheel feature and vehicle speed feature, so as to obtain the steering intention of the user.

[0085] That is, the full connection layer is used to flatten the feature sequence output by the pooling layer into a one-dimensional vector, and the feature fusion and classification are performed through the full connection layer. The full connection layer output dimension is the number of steering intention categories, such as left turn, right turn, straight driving, etc.

[0086] The input data can be represented in the form of a vector:

[0087] T=[Aleft,Aright,B,C]

[0088] Aleft and Aright represent the attention features of the driver's line of sight on the left and right sides, respectively, B represents the steering angle feature of the steering wheel, and C represents the vehicle speed feature.

[0089] Activation function: an activation function is added after the convolution layer and the full connection layer, a nonlinear factor is introduced, and the expression ability of the model is enhanced.

[0090] The scheme adopts a deep learning algorithm to establish a steering decision model, which can adapt to different users through feature extraction, feature fusion, classification prediction and other processes, and has strong pertinence and good adaptability. The deep learning algorithm can learn the preferences and behavior patterns of users, so as to more accurately predict the next operation of the user, and thus ensure the prediction accuracy of the steering intention of the user and provide good automatic starting service of the turn signal.

[0091] In this embodiment, the steering decision model is also subjected to model training, and the model training includes:

[0092] Driving data is collected and data labeling is performed, and in the data labeling, the line-of-sight feature, steering wheel feature and vehicle speed feature and the corresponding steering intention label are included.

[0093] The driving data is divided into a training set, a validation set and a test set according to a preset ratio; wherein the preset ratio is set according to actual needs, for example, 7:2:1.

[0094] The steering decision model is trained using the training set data, and the model parameters θ are optimized through a backpropagation algorithm to minimize the loss function;

[0095] The trained steering decision model is evaluated using the test set, and the evaluation criteria include one or more of accuracy, recall rate, and F1 value to ensure that the model has good generalization ability.

[0096] This solution collects a large amount of annotated driving data for training, and the availability of big data enables the model to learn underlying patterns and reduce overfitting, thereby enabling it to handle more complex input data and task scenarios, achieving higher prediction accuracy and classification effectiveness.

[0097] The real-time collected user state information and the vehicle state are input into the trained steering decision model, and the model outputs the corresponding steering intent prediction result:

[0098] X = f(T; θ)

[0099] Where X is the predicted steering intent, f is the steering decision model, T is the input data, and θ is the model parameter.

[0100] In this embodiment, Kalman filter, Bayesian network or other fusion algorithms can be used according to needs. Through a large amount of experimental data, the algorithm is optimized, and the algorithm parameters are adjusted. These algorithms can dynamically adjust their weights in the final result according to the reliability and correlation of different data, improving the accuracy and reliability of steering intent recognition. For example, to meet the demand of complex scenarios, a classifier such as decision tree or neural network can be trained using machine learning or deep learning algorithms to distinguish different steering intents.

[0101] In this embodiment, after inputting the user state information and the vehicle state information into the steering decision model, feature extraction is performed first, and the obtained features are as follows:

[0102] Gaze feature: extract features such as the time the gaze stays in the rearview mirror or the front direction, the speed and direction of the gaze movement, and record this information A(D SightLine ).

[0103] Steering wheel feature: extract features such as the size and rate of change of the steering wheel angle, and record this information B(D SteeringWheel ).

[0104] Speed feature: extract features such as the trend of the vehicle speed, and record this information C(D VehicleSpeed ).

[0105] In this embodiment, when the fusion algorithm is a Bayesian network, a feature fusion process is performed according to the line-of-sight feature, the steering wheel feature and the vehicle speed feature, and a classification prediction is performed to obtain the steering intention of the user, including:

[0106] b1, performing scene classification based on different driving scenes to set corresponding steering intentions;

[0107] Wherein, the driving scene includes a left lane change scene, a right lane change scene, a left U-turn scene, a right U-turn scene and a straight driving scene. The left lane change scene and the left U-turn scene correspond to the "left" steering intention, the right lane change scene and the right U-turn scene correspond to the "right" steering intention, and the straight driving scene corresponds to the "non" steering intention.

[0108] b2, taking the line-of-sight feature, the steering wheel feature and the vehicle speed feature obtained according to the user state information and the vehicle state information as observation variables, and combining the Bayesian theorem to calculate the prior probability corresponding to each steering intention. The calculation formula of the prior probability is as follows:

[0109]

[0110] Wherein, the line-of-sight feature A (such as line-of-sight angle), the steering wheel feature B (such as steering wheel turning angle) and the vehicle speed feature C are observation variables, and X is the steering intention. This formula represents the probability of calculating the steering intention X given the observation values A, B and C. P(X) is the prior probability of the steering intention X, which can be obtained by statistically analyzing the frequency of different steering intentions (such as left turn, right turn and straight driving) in a large number of driving data. For example, if left turn occurs n1 times in N driving records, then

[0111] Wherein, the Bayesian network is a probabilistic graphical model used to represent the dependency relationship between variables.

[0112] P(A,B,C|X) is the joint conditional probability of the line-of-sight feature A, the steering wheel feature B and the vehicle speed feature C given the steering intention X. This needs to be learned from a large amount of driving data. Assuming that the line-of-sight feature A can be discretized as [a1, a2, …, am], the steering wheel feature B can be discretized as [b1, b2, …, bn], the vehicle speed feature C can be discretized as [c1, c2, …, co], and the steering intention X can be discretized as [x1, x2, …, xp]. Then P(A=ai,B=bj,C=ck|X=xl) represents the probability of line-of-sight angle ai, steering wheel turning angle bj and vehicle speed ck when the steering intention is xl.

[0113] P(A,B,C) is the joint probability of the line-of-sight angle A, the steering wheel turning angle B and the vehicle speed C, which can be represented by the total probability formula:

[0114] P(A,B,C)=∑ x∈x P(A,B,C|x)P(x)

[0115] b3. Compare the prior probabilities of all the stated turning intentions, and output the turning intention with the highest probability value as the classification prediction result.

[0116] This solution incorporates fusion algorithms, such as Bayesian networks and Kalman filters, into a steering decision model based on deep learning algorithms. When extracting gaze features, steering wheel features, and vehicle speed features, it considers the probabilistic relationships between various factors based on the reliability and relevance of different data, dynamically adjusts their weights in the final result, and then uses a probabilistic model to characterize and predict different user steering intentions, thereby providing more reliable prediction results and improving the accuracy and reliability of steering intention recognition.

[0117] S3. Confirm whether to execute automatic turn signal activation control based on the user's stated steering intention, including:

[0118] The user's steering intention and corresponding prior probability are obtained from the steering decision model. It is determined whether the prior probability is greater than a preset threshold. If so, the automatic start control of the turn signal is confirmed. Then, a corresponding control command is generated based on the steering intention to drive the corresponding left / right turn signal to start.

[0119] Specifically, a probability threshold δ is set as a preset threshold. When the probability of a certain steering intention predicted by the steering decision model is greater than δ, it is determined that the driver has that steering intention. If the driving behavior and environmental conditions for left and right turns are similar, the same feature value threshold δ can be used, and δ = δleft = δright can be set. If the experimental data shows that there are significant differences in driving behavior for left and right turns, different thresholds δleft and δright can be set.

[0120] Therefore, the conditions for turning on the turn signal are as follows:

[0121] Turn on the left turn signal, if Xleft>δleft

[0122] Turn on the right turn signal, if Xright>δright

[0123] Without turning on the turn signal, if X < δ

[0124] Where Xleft and Xright are the probabilities of the model's predicted left and right turn intentions, respectively.

[0125] When the steering light opening condition is determined, the intention is sent as an in-vehicle signal to a vehicle control responsible steering light control module; the steering light control module receives the signal, determines that the current steering light is not on, and automatically turns on the steering light on the corresponding side (i.e. the side of the driver's line of sight and the steering wheel with the intention of turning).

[0126] The present scheme sets a probability threshold as a preset threshold based on the actual application environment, compares whether the prior probability of the output steering intention is greater than the preset threshold, judges whether to execute the steering light self-starting control, thereby reducing the probability of steering light false triggering and improving driving safety.

[0127] The present application also provides a vehicle steering light self-starting system, comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor executes the computer program to realize the vehicle steering light self-starting method as described above.

[0128] In the embodiment, a data acquisition module connected with the processor is further included, and the data acquisition module comprises an image acquisition module, a vehicle speed acquisition module and a steering angle acquisition unit.

[0129] The image acquisition module is used for acquiring user state information in real time.

[0130] The vehicle speed acquisition unit is used for acquiring vehicle speed data.

[0131] The steering angle acquisition unit is used for acquiring steering wheel data of the vehicle.

[0132] The present scheme defines the image acquisition module, the vehicle speed acquisition module and the steering angle acquisition unit for data acquisition based on the driver monitoring system and the vehicle body domain control system of the vehicle itself, without the need to install additional equipment, and is applicable to different types of vehicles and has good applicability.

[0133] Specifically, the steering light self-starting working principle of the present application is as follows:

[0134] The driver monitoring system (DMS) has a line-of-sight interaction function. During driving, the driver's line-of-sight attention direction is identified in real time. When the driver's line-of-sight attention is detected to be on the corresponding left or right side rearview mirror and the left or right front side area, it may imply that the driver has the intention to turn or change lanes to that side. At the same time, the invention further monitors other turning information of the vehicle as an auxiliary. If the steering wheel is detected to have the intention to turn to the corresponding side (such as by monitoring the steering wheel turning angle) during the line-of-sight attention to the corresponding side, and combined with the current speed change trend, the corresponding side turn signal will be automatically turned on, thereby realizing the automatic turn signal function. The invention does not require the driver to manually operate the turn signal, improves the automation level of turn signal operation, and helps to reduce traffic safety problems caused by improper turn signal operation.

[0135] The beneficial effects of the invention are as follows:

[0136] 1. Improve the automation level of turn signal operation

[0137] The invention can automatically turn on the turn signal according to the driver's line-of-sight direction intention through the cooperative work of the driver monitoring system (DMS) and the turning monitoring (vehicle speed data and steering wheel data). This greatly reduces the steps of manual operation, improves the convenience and automation level of operation.

[0138] 2. Enhance traffic safety

[0139] Due to the reduction of manual operation of the turn signal, the situation of not turning on the turn signal due to the driver's forgetfulness or untimely operation is avoided. When the vehicle turns or changes lanes, turning on the turn signal correctly is crucial for conveying the driving intention to other road users. The technology can effectively reduce traffic accidents caused by improper turn signal operation and ensure road traffic safety.

[0140] 3. Improve the comfort of driving experience

[0141] The driver does not need to be distracted to manually operate the turn signal, especially in complex road conditions that require frequent turning or lane changing. The driver can focus more on driving operations such as controlling speed and maintaining distance. By realizing automatic turn signal operation to reduce the driver's operation burden, the comfort of driving experience is improved.

[0142] 4. Improve the accuracy of turning intention recognition

[0143] Various data fusion algorithms, such as Kalman filter, Bayesian network and other fusion algorithms, can be used to dynamically adjust the weight according to the reliability and correlation of different data. In complex scenarios, machine learning or deep learning algorithms can also be used to train classifiers (such as support vector machines, decision trees or neural networks) to distinguish different steering intentions. This effectively improves the accuracy of driver steering intention recognition, reduces the possibility of misjudgment, and further improves the reliability of the automatic turn signal function.

[0144] The above embodiments are the preferred embodiments of the present application, but the embodiments of the present application are not limited to the above embodiments, and any changes, modifications, substitutions, combinations, simplifications made without departing from the spirit and principles of the present application shall be equivalent replacement methods and shall be within the scope of protection of the present application.

Claims

1. A method for automatically starting a vehicle turn signal, characterized in that, include: Real-time acquisition of user status information based on driver monitoring system; Real-time collection of vehicle status information; A steering decision model is constructed based on a deep learning algorithm. The user state information and the vehicle state information are input into the steering decision model to perform feature fusion processing and classification prediction in order to obtain the user's steering intention. Confirm whether to execute automatic turn signal control based on the user's stated steering intention; Based on the driver monitoring system, user status information is obtained in real time, including: Divide the space around the vehicle into zones to obtain target zones corresponding to different steering intentions; The driver monitoring system collects the user's head image in real time, performs image recognition to track the gaze, and performs region matching with the target area to determine the user's gaze information. The user's gaze information includes one or more of the following: gaze angle, gaze dwell time, gaze movement speed, and gaze movement direction; It also includes: using an infrared sensor to perform pupil localization and corneal reflective point detection to assist image recognition in achieving gaze tracking; and / or, acquiring the user's head posture and using a deep learning algorithm to estimate the user's head's three-axis posture, and then combining the three-axis posture to dynamically compensate for gaze tracking achieved by image recognition; The steering decision model includes an input layer, a convolutional layer, a pooling layer, a fully connected layer, and an activation function. The input layer is used to normalize the user status information and the vehicle status information to obtain corresponding input features; The convolutional layer is used to perform convolution operations on the input features through several convolutional kernels in order to extract the corresponding gaze features, steering wheel features and vehicle speed features. The pooling layer is used to downsample the gaze features, steering wheel features, and vehicle speed features to reduce the feature dimensionality and output the corresponding feature sequence. The fully connected layer has a built-in fusion algorithm, which is used to perform feature fusion processing and classification prediction based on the gaze features, steering wheel features and vehicle speed features to obtain the user's steering intention.

2. The method for automatically starting a vehicle turn signal as described in claim 1, characterized in that, Real-time collection of vehicle status information involves: continuously monitoring the vehicle using the vehicle body domain control system and obtaining corresponding vehicle body information; and extracting vehicle speed data and steering wheel data from the vehicle body information as the current vehicle status information.

3. The method for automatically starting a vehicle turn signal as described in claim 1, characterized in that, When the fusion algorithm is a Bayesian network, feature fusion processing and classification prediction are performed based on the gaze features, steering wheel features, and vehicle speed features to obtain the user's steering intention, including: Different driving scenarios are categorized to set corresponding steering intentions; The line-of-sight features, steering wheel features, and vehicle speed features obtained based on the user status information and the vehicle status information are used as observation variables, and the prior probability corresponding to each steering intention is calculated by combining Bayes' theorem. Compare the prior probabilities of all stated turning intentions, and output the turning intention with the highest probability value as the classification prediction result.

4. The method for automatically starting a vehicle turn signal as described in claim 3, characterized in that, Based on the user's stated steering intention, determine whether to execute automatic turn signal activation control, including: The user's steering intention and corresponding prior probability are obtained from the steering decision model. It is determined whether the prior probability is greater than a preset threshold. If so, the automatic start control of the turn signal is confirmed. Then, a corresponding control command is generated based on the steering intention to drive the corresponding left / right turn signal to start.

5. A method for automatically starting a vehicle turn signal as described in claim 3, characterized in that: It also includes training a steering decision model, wherein the model training includes: Driving data is collected and labeled, including gaze characteristics, steering wheel characteristics, vehicle speed characteristics, and corresponding steering intention labels. The driving data is divided into training set, validation set and test set according to a preset ratio; The steering decision model is trained using the training set data, and the model parameters θ are optimized and the loss function is minimized using the backpropagation algorithm. The trained steering decision model is evaluated using a test set, and the evaluation criteria include one or more of the following metrics: accuracy, recall, and F1 score.

6. A vehicle turn signal automatic start system, characterized in that: The device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, when the processor executes the computer program, it implements a vehicle turn signal self-starting method as described in any one of claims 1-5.

7. A vehicle turn signal self-starting system as described in claim 6, characterized in that: It also includes a data acquisition module connected to the processor, the data acquisition module including an image acquisition module, a vehicle speed acquisition module and a corner acquisition unit; The image acquisition module is used to collect user status information in real time; Vehicle speed acquisition unit, used to collect vehicle speed data; The corner acquisition unit is used to collect steering wheel data of the vehicle.

Images