Driver behavior monitoring and analyzing method based on artificial intelligence

Abstract

The invention relates to the technical field of data processing, in particular to a driver behavior monitoring analysis method based on artificial intelligence, and the method comprises the steps: carrying out the behavior triggering analysis of at least two behavior atoms according to driver feature information and vehicle dynamic information in a vehicle driving process of a driver, and determining a behavior monitoring moment; monitoring the behavior state change of each behavior atom within a preset duration after the behavior monitoring moment, carrying out behavior chain matching on the driver, and if the matching is successful, adjusting the basic weight of any behavior atom according to the real-time lane departure distance, the real-time vehicle speed and the behavior triggering time difference of any behavior atom, so as to obtain the behavior state change of any behavior atom. Obtaining a space-time attenuation weight; the behavior composite risk value is obtained by combining the space-time attenuation weight and the behavior triggering characteristic value of each behavior atom, and abnormal early warning is performed on the driver behavior according to the behavior composite risk value, so that the accuracy of driver behavior monitoring based on a single threshold is improved.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a driver behavior monitoring and analysis method based on artificial intelligence. Background Art

[0002] In recent years, with the widespread adoption of intelligent driving technology and the strengthening of road safety regulations, driver behavior monitoring has become a core module of vehicle active safety systems. According to statistics, approximately 70% of traffic accidents worldwide are directly related to driver distraction, fatigue, and dangerous driving behaviors.

[0003] Traditional driver behavior monitoring methods mainly rely on on-board sensors (such as steering wheel angle monitoring) or rule-based visual analysis (such as eye tracking and head posture detection). However, there are a series of problems in practical applications. Traditional methods are based on preset thresholds (such as the duration of eye closure) or fixed behavior templates, and cannot dynamically adapt to different driving scenarios (such as urban congestion and high-speed cruising) and drivers' personalized habits. The recognition accuracy of "hidden risk behaviors" (such as brief distraction and emotional fluctuations) is low, and single behaviors (such as fatigue or distraction) are analyzed in isolation. There is a lack of temporal correlation modeling of "behavior chains" (such as continuous yawning accompanied by lane deviation), and dynamic assessment of complex risks cannot be achieved. Summary of the Invention

[0004] In view of this, an embodiment of the present invention provides a driver behavior monitoring and analysis method based on artificial intelligence to solve the problem that dynamic assessment of compound risks cannot be achieved when monitoring driver behavior.

[0005] An embodiment of the present invention provides a driver behavior monitoring and analysis method based on artificial intelligence, the method comprising the following steps: While the driver is driving the vehicle, driver characteristic information and vehicle dynamic information are collected and detected. Based on the driver characteristic information and vehicle dynamic information, behavior trigger analysis is performed on at least two behavior atoms to determine the behavior monitoring time. The driver characteristic information includes the duration of continuous eye closure, head deflection angle, and steering wheel grip. The vehicle dynamic parameters include steering wheel angle, vehicle speed, and lane departure distance. Monitor the behavior state changes of each behavior atom within a preset time period after the behavior monitoring moment, perform behavior chain matching on the driver based on the behavior state changes of each behavior atom, and if the match is successful, obtain the real-time lane departure distance, real-time vehicle speed, and the behavior triggering time difference and behavior triggering characteristic value of each behavior atom; For any behavior atom, a basic weight of the behavior atom is obtained according to the driving scenario, and the basic weight of the behavior atom is adjusted according to the real-time lane departure distance, the real-time vehicle speed, and the behavior triggering time difference of the behavior atom to obtain the spatiotemporal attenuation weight of the behavior atom; The spatiotemporal attenuation weight of each behavior atom is obtained, and the composite risk value of the driver's behavior is obtained by combining the spatiotemporal attenuation weight of each behavior atom with the behavior triggering characteristic value. An abnormal warning of the driver's behavior is issued based on the composite risk value of the behavior.

[0006] Preferably, the behavior atoms include a continuous eye closure behavior atom, a lane departure behavior atom, a steering wheel angle high volatility behavior atom, a grip force behavior atom, and a head deviation behavior atom. Based on the driver characteristic information and the vehicle dynamic information, a behavior trigger analysis is performed on at least two behavior atoms to determine the behavior monitoring time, including: If it is detected that the driver's continuous eye closure time is greater than or equal to the preset eye closure time, the continuous eye closure behavior atom is triggered; if it is detected that the lane deviation distance is less than or equal to the preset ratio of the lane width and lasts for a first preset time, the lane departure behavior atom is triggered; if it is detected that the standard deviation of the steering wheel angle change rate within a preset time period is greater than a preset change rate standard deviation threshold, the steering wheel angle high volatility behavior atom is triggered; if it is detected that the standard deviation of the steering wheel grip force within a preset time period is less than a preset grip force standard deviation threshold, the grip force behavior atom is triggered; if it is detected that the driver's head deflection angle is greater than a preset angle and lasts for a second preset time period, the head deviation behavior atom is triggered; The time point corresponding to the first triggered behavior atom is determined, and the time point is used as the behavior monitoring moment.

[0007] Preferably, the performing behavior chain matching on the driver according to the behavior state change of each behavior atom includes: If the continuous closing of eyes behavior atom is triggered continuously within the preset time period, and there is a lane departure behavior atom triggered, and the vehicle speed is greater than the preset speed, then it is determined that the driver meets the fatigue-loss of control behavior chain; If the head deviation behavior atom is triggered continuously within a preset time period and the vehicle's deceleration is greater than the emergency braking deceleration threshold, the driver is determined to have met the distraction-misoperation behavior chain; If the standard deviation of the steering wheel grip force within the preset time period is less than the preset grip force standard deviation threshold, and the standard deviation of the steering wheel angle change rate is greater than the preset change rate standard deviation threshold, it is determined that the driver meets the physiological abnormality-out-of-control behavior chain.

[0008] Preferably, the method for obtaining the behavior triggering time difference and the behavior triggering characteristic value of each behavior atom includes: For any behavior atom, determine the time point when any behavior atom is triggered for the first time within the preset time length, record it as the initial triggering moment, calculate the time interval between the initial triggering moment and the last moment within the preset time length, record it as the behavior triggering time difference of any behavior atom, if the behavior triggering time difference of any behavior atom is not 0, set the behavior triggering characteristic value of any behavior atom to the first preset value, if the behavior triggering time difference of any behavior atom is 0, set the behavior triggering characteristic value of any behavior atom to the second preset value.

[0009] Preferably, the driving scenario includes a highway scenario and an urban scenario, and the step of obtaining the basic weight of any behavior atom according to the driving scenario includes: If any of the behavior atoms is a lane departure behavior atom, the basic weight of the behavior atom in the highway scenario is set to a first parameter value, and the basic weight in the urban scenario is set to a second parameter value, and the first parameter value is greater than the second parameter value; If any of the behavior atoms is a head shift behavior atom, the basic weight of the behavior atom in the urban scene is set to the second parameter value, the number of high-risk events in which the behavior atom participated and the total number of high-risk events in the high-speed scene during the historical period are counted, and the ratio of the number of high-risk events in which the behavior atom participated to the total number of high-risk events is calculated as the basic weight of the behavior atom in the high-speed scene; If any of the behavior atoms is not a lane departure behavior atom and a head deviation behavior atom, then the number of high-risk events in which any of the behavior atoms participates and the total number of high-risk events in any scenario during the historical period are counted, and the ratio of the number of high-risk events in which any of the behavior atoms participates to the total number of high-risk events is calculated as the basic weight of any of the behavior atoms in any scenario; wherein, when the vehicle speed is greater than the preset speed, the basic weight of the grip behavior atom in the urban scenario is set to the third parameter value.

[0010] Preferably, adjusting the basic weight of any behavior atom according to the real-time lane departure distance, the real-time vehicle speed, and the behavior triggering time difference of any behavior atom to obtain the spatiotemporal attenuation weight of any behavior atom includes: Adaptively obtaining a time decay coefficient according to the real-time vehicle speed, and using the inverse of the product of the behavior triggering time difference of any behavior atom and the time decay coefficient as the independent variable of an exponential function with a natural constant as the base to obtain a time adjustment coefficient; Calculate the ratio of the real-time lane departure distance to the lane width, and obtain the difference between the constant 1 and the ratio, which is recorded as the spatial adjustment coefficient; The product of the basic weight of any behavior atom, the time adjustment coefficient and the space adjustment coefficient is used as the spatiotemporal attenuation weight of any behavior atom.

[0011] Preferably, the adaptively obtaining the time attenuation coefficient according to the real-time vehicle speed includes: Calculate the speed ratio between the real-time vehicle speed and the preset maximum speed, obtain the product between the preset speed sensitivity coefficient and the speed ratio, and use the sum of the constant 1 and the product as the coefficient adjustment value; and use the product of the basic attenuation rate and the coefficient adjustment value as the time attenuation coefficient.

[0012] Preferably, the combining of the spatiotemporal attenuation weights of each behavior atom and the behavior triggering characteristic value to obtain the driver's behavior composite risk value includes: Based on the spatiotemporal attenuation weight of each behavior atom, the behavior triggering feature values ​​of all behavior atoms are weighted summed to obtain the driver's behavior composite risk value.

[0013] Preferably, the abnormal warning of the driver's behavior based on the composite risk value of the behavior includes: Obtain the preset first risk warning value and second risk warning value, and the first risk warning value is greater than the second risk warning value. If the composite risk value of the behavior is greater than the first risk warning value and less than the second risk warning value, a voice prompt is given; if the composite risk value of the behavior is greater than or equal to the second risk warning value, emergency braking is performed.

[0014] Compared with the prior art, the embodiments of the present invention have the following beneficial effects: The present invention acquires multi-source data, defines basic behavioral atoms, and obtains the risk weight (basic weight) associated with each behavior. According to the triggering situation of each behavioral atom, the driver's driving behavior is detected by behavioral chain activation. After matching the behavioral chain, the spatiotemporal attenuation weight of each behavioral atom is dynamically obtained according to the corresponding behavioral chain. Then, by superimposing the weights, the driver's behavioral composite risk value is obtained, which is used for warnings of different degrees. This enhances the generalization ability of behavior monitoring and improves the accuracy of driver behavior monitoring based only on a single threshold. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0016] Figure 1This is a method flow chart of an artificial intelligence-based driver behavior monitoring and analysis method provided in Example 1 of the present invention. DETAILED DESCRIPTION

[0017] The embodiments of the present disclosure are described in detail below, and examples of the embodiments are shown in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and intended to be used to explain the present disclosure, but should not be understood as limiting the present disclosure.

[0018] It should be noted that the terms "first," "second," and the like in the specification of the present disclosure and the accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or precedence. It should be understood that the terms used in this manner are interchangeable where appropriate, so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present disclosure. Instead, they are merely examples of devices and methods consistent with some aspects of the present disclosure.

[0019] In order to illustrate the technical solution of the present invention, specific embodiments are provided below.

[0020] See also Figure 1 , is a method flow chart of a driver behavior monitoring and analysis method based on artificial intelligence provided by the first embodiment of the present invention, such as Figure 1 As shown, the method may include: Step S101 : While the driver is driving the vehicle, the driver's characteristic information and the vehicle's dynamic information are collected and detected, and behavior triggering analysis is performed on at least two behavior atoms based on the driver's characteristic information and the vehicle's dynamic information to determine the behavior monitoring time.

[0021] The present invention primarily monitors the driver's behavior while driving a vehicle. Driver behavior is primarily characterized by head posture, eye state, and physiological changes. Therefore, while the driver is driving, the system primarily monitors and collects visual behavioral characteristics, vehicle dynamic parameters, and driver physiological information. Specifically, an infrared wide-angle camera is mounted behind the rearview mirror in the center of the instrument panel to collect the driver's eye closure and head posture information. PERCLOS (Percentage of Eye Closure) outputs the duration of eye closure, and the head deflection angle is output based on the posture. This is prior art and will not be described in detail here. A capacitive grip sensor embedded in the steering wheel at the 3 / 9 o'clock position collects the driver's steering wheel grip to characterize the driver's physiological information. Various sensors deployed on the vehicle collect information such as steering wheel angle, vehicle speed, and lane departure distance for characterization of vehicle dynamic parameters. Lane departure distance refers to the distance between either side of the vehicle and the lane line.

[0022] Based on the above data, the duration of eye closure, head deflection angle, and steering wheel grip force are combined to form driver characteristic information, while the steering wheel angle, vehicle speed, and lane departure distance are combined to form vehicle dynamic parameters. In this embodiment of the present invention, while the driver is driving the vehicle, real-time collection and detection of driver characteristic information and vehicle dynamic information are performed to perform behavior trigger analysis on at least two behavior atoms based on the driver characteristic information and vehicle dynamic information to determine the behavior monitoring time.

[0023] Specifically, due to the different operating habits of different drivers, there may be certain differences in the corresponding driving behaviors. In order to enhance the generalization ability and increase the risk assessment accuracy of driving behavior monitoring based only on a single threshold, the embodiment of the present invention sets behavior atoms including continuous eye closure behavior atoms, lane departure behavior atoms, steering wheel angle high volatility behavior atoms, grip behavior atoms and head deviation behavior atoms.

[0024] When the driver is driving the vehicle, if it is detected that the driver's eyes are closed for more than or equal to 2 seconds (that is, the eyes are closed for 2 consecutive seconds), the continuous eyes closing behavior atom is triggered; if the lane deviation distance is less than or equal to 10%-15% of the lane width and lasts for at least 1 second, the lane departure behavior atom is triggered; according to the acquisition frequency of 50HZ, that is, one data point is collected every 0.02 seconds, 500 steering wheel angles collected in 10 seconds are obtained, and the angle fluctuation rate between two adjacent steering wheel angles is obtained. , represents the i-th steering wheel angle, represents the i-1th steering wheel angle, then the standard deviation of the steering wheel angle fluctuation rate between each two adjacent steering wheel angles in the 500 steering wheel angles is calculated to obtain the standard deviation of the steering wheel angle change rate. If the standard deviation of the steering wheel angle change rate detected within 10 seconds is greater than the preset change rate standard deviation threshold (experienced set to ), the steering wheel angle high volatility behavior atom is triggered; similarly, according to the acquisition frequency of 50HZ, that is, one data point is collected every 0.02 seconds, 500 steering wheel grip forces collected within 10 seconds are obtained, and the standard deviation of the 500 steering wheel grip forces is calculated. If the standard deviation of the steering wheel grip forces within 10 seconds is less than the preset grip standard deviation threshold (set empirically as ), the grip behavior atom is triggered; if it is detected that the driver's head deflection angle (left or right) is greater than the preset angle of 15 degrees and lasts for at least 5 seconds, the head deviation behavior atom is triggered.

[0025] When any behavior atom is triggered for the first time, behavior monitoring is performed with the triggering moment as the time node, that is, the time point corresponding to the first triggered behavior atom is used as the behavior monitoring moment to start abnormal monitoring of the driver's driving behavior.

[0026] Step S102 monitors the behavior state changes of each behavior atom within a preset time period after the behavior monitoring moment, and performs behavior chain matching on the driver based on the behavior state changes of each behavior atom. If the match is successful, the real-time lane departure distance, real-time vehicle speed, and the behavior triggering time difference and behavior triggering characteristic value of each behavior atom are obtained.

[0027] After determining the behavior monitoring moment, all behavior atoms within the subsequent period are monitored. The monitoring duration for all behavior atoms is set to 30 seconds after the behavior monitoring moment, that is, the preset duration is 30 seconds. If multiple behavior atoms are triggered, the driver's behavior chain matching is performed. The embodiment of the present invention only lists typical behavior chains, and other behavior chains can be expanded according to actual conditions. Among them, behavior chain 1: fatigue-loss of control chain, lane deviation occurs after closing the eyes for a certain period of time, and the vehicle speed is greater than a certain speed threshold; behavior chain 2: distraction-misoperation chain, sudden braking after the head continuously deviates by a certain angle; behavior chain 3: physiological abnormality-loss of control chain, both grip strength changes and steering wheel angle fluctuation rate increase suddenly.

[0028] Therefore, if the continuous eye closure behavior atom is continuously triggered within the preset time length, and there is a triggered lane departure behavior atom, and the vehicle speed is greater than the preset speed, then the driver is determined to meet the fatigue-loss of control behavior chain; if the head deviation behavior atom is continuously triggered within the preset time length, and the vehicle's deceleration is greater than the emergency braking deceleration threshold, then the driver is determined to meet the distraction-misoperation behavior chain; if the standard deviation of the steering wheel grip force within the preset time length is less than the preset grip force standard deviation threshold, and the standard deviation of the steering wheel angle change rate is greater than the preset change rate standard deviation threshold, then the driver is determined to meet the physiological abnormality-loss of control behavior chain.

[0029] In one embodiment, if the driver closes his eyes for 2 consecutive seconds and then deviates from the lane, and the vehicle speed is greater than 80 km / h within 30 seconds after the behavior monitoring moment, the driver is determined to have met the fatigue-loss of control behavior chain; if the driver brakes suddenly within 3 seconds after the head deviates continuously by 15 degrees within 30 seconds after the behavior monitoring moment, the driver is determined to have met the distraction-misoperation behavior chain, where the judgment method for sudden braking is: the deceleration is greater than 0.4g, where g is the acceleration due to gravity , that is, the deceleration is greater than 3.92m / s. 0.4g is often the deceleration threshold for emergency braking, which is common in collision avoidance or sudden danger; if the standard deviation of the driver's steering wheel grip within 30 seconds after the behavior monitoring time is less than 0.5N, and the standard deviation of the steering wheel angle fluctuation rate is greater than , it is determined that the driver meets the physiological abnormality-out-of-control behavior chain.

[0030] Based on the above-mentioned behavior chain matching method, if the behavior chain matching is successful, the corresponding behavior composite risk value is calculated. If the behavior chain matching is unsuccessful, the monitoring value under a single behavior atom is compared with the historical experience threshold to perform a single-dimensional hard threshold warning.

[0031] For example: if the driver's eyes are detected to be closed at the 30th second and the eyes are closed continuously for 32 seconds, the continuous eyes closing behavior atom is triggered. At this time, all data within 30 seconds after the 30th second is monitored. For example, lane deviation occurs at the 35th second and the vehicle speed reaches 85km / h at the 40th second. At this time, the behavior chain matches behavior chain 1. After the behavior chain matches successfully, the behavior composite risk calculation can be performed subsequently. If the driver's eyes are detected to be closed at the 30th second and the eyes are closed continuously for 32 seconds, and no other behavior atoms are triggered in the subsequent 30 seconds, the behavior chain match fails. That is, the judgment is made based on a single behavior atom. Since only the continuous eyes closing behavior atom is activated, the international standard for mild fatigue is PERCLOS greater than or equal to 15%, that is, eyes closed for 4.5 seconds in 30 seconds. However, the eyes are closed only from the 30th to the 32nd second, and the rest are normal. It does not meet the requirements of mild fatigue and no abnormal warning is required.

[0032] Therefore, after successfully matching the driver's behavior chain, we can obtain the real-time lane departure distance d (that is, the distance between the vehicle and the nearest lane line), the real-time vehicle speed v, and the behavior triggering time difference of each behavior atom 30 seconds after the behavior monitoring moment. and the behavior triggering characteristic value I, which is used to analyze the composite risk of the driver's behavior.

[0033] The method for obtaining the behavior triggering time difference and behavior triggering characteristic value of each behavior atom includes: For any behavior atom, determine the time point when any behavior atom is triggered for the first time within the preset time length, record it as the initial triggering moment, calculate the time interval between the initial triggering moment and the last moment within the preset time length, record it as the behavior triggering time difference of any behavior atom, if the behavior triggering time difference of any behavior atom is not 0, set the behavior triggering characteristic value of any behavior atom to the first preset value, that is, the constant 1, if the behavior triggering time difference of any behavior atom is 0, set the behavior triggering characteristic value of any behavior atom to the second preset value, that is, 0.

[0034] In step S103, for any behavior atom, a basic weight of any behavior atom is obtained according to the driving scenario, and the basic weight of any behavior atom is adjusted according to the real-time lane departure distance, the real-time vehicle speed, and the behavior triggering time difference of any behavior atom to obtain the spatiotemporal attenuation weight of any behavior atom.

[0035] A basic weight is set for each behavior atom based on its risk contribution in different scenarios. Vehicle driving scenarios are primarily categorized as highway and urban, and the basic weights vary across scenarios. Therefore, for any behavior atom, if it is a lane departure atom, the basic weight for the lane departure atom is 1.2 in highway scenarios and 0.5 in urban scenarios, given that the risk of vehicle deviation in highway scenarios is far greater than in urban areas.

[0036] In urban scenarios, considering the existence of congestion, there will be frequent observation of the surroundings. Therefore, if any of the behavior atoms is a head offset behavior atom, the basic weight of the head offset behavior atom in the urban scenario is set to 0.5. Then, for the basic weight in the high-speed scenario, the number of high-risk events and the total number of high-risk events in which the head offset behavior atom participated in the high-speed scenario during the historical period are counted, and the ratio of the number of high-risk events in which the head offset behavior atom participated to the total number of high-risk events is calculated as the basic weight of the head offset behavior atom in the high-speed scenario.

[0037] If any of the behavior atoms is not a lane departure behavior atom or a head deviation behavior atom, then the number of high-risk events in which any of the behavior atoms participated and the total number of high-risk events in any scenario during the historical period are counted, and the ratio of the number of high-risk events in which any of the behavior atoms participated to the total number of high-risk events is calculated as the basic weight of any of the behavior atoms in any scenario.

[0038] It is worth noting that the faster the vehicle speed, the greater the danger of an emergency. Therefore, when the vehicle speed is greater than 60 km / h, the basic weight of the grip behavior atom in the urban scene is set to 1.5.

[0039] Therefore, based on the above analysis, the embodiment of the present invention can obtain the basic weight of each behavior atom according to the driving scenario of the driver, and then adjust and update the basic weight of any behavior atom according to the real-time lane departure distance, the real-time vehicle speed and the behavior triggering time difference of any behavior atom to obtain an adaptive weight that conforms to the real-time driving state, that is, the spatiotemporal attenuation weight of any behavior atom. The spatiotemporal attenuation weight of any behavior atom is obtained as follows: (1) Calculate the speed ratio between the real-time vehicle speed and the preset maximum speed, obtain the product of the preset speed sensitivity coefficient and the speed ratio, and use the sum of the constant 1 and the product as the coefficient adjustment value; and use the product of the basic attenuation rate and the coefficient adjustment value as the time attenuation coefficient.

[0040] The time decay coefficient is determined based on the reaction time of most drivers. 95% of drivers correct lane departure within 5 seconds, so the time decay coefficient is set , so that the weight decays to 60% of the initial value within 5 seconds. However, since the speed of the vehicle will affect the driver's reaction time, the time decay coefficient can be adaptively updated according to the real-time speed. The updating method is:

[0041] in, represents the time decay coefficient, Represents the basic decay rate, taking , Indicates real-time speed, Indicates the maximum speed of the vehicle. Since the speed limit on the highway is 120km / h, Pick , 1 represents a constant.

[0042] (2) The opposite of the product of the behavior triggering time difference of any behavior atom and the time attenuation coefficient is used as the independent variable of the exponential function with the natural constant as the base to obtain the time adjustment coefficient Calculate the ratio of the real-time lane departure distance to the lane width, and obtain the difference between the constant 1 and the ratio, which is recorded as the spatial adjustment coefficient The product of the basic weight of any behavior atom, the time adjustment coefficient and the space adjustment coefficient is used as the spatiotemporal attenuation weight of any behavior atom.

[0043] The calculation formula for the spatiotemporal attenuation weight of any behavioral atom is:

[0044] in, represents the spatiotemporal attenuation weight of the i-th behavior atom, represents the basic weight of the i-th behavior atom, e represents a natural constant, represents the time decay coefficient, Indicates the behavior triggering time difference of the i-th behavior atom, Indicates the real-time lane departure distance, Indicates lane width, 1 indicates a constant.

[0045] It should be noted that since the risks of driver behavior are time-sensitive, for example, the risk of closing one's eyes for 2 seconds will decrease after the occurrence as the driver may self-correct (such as refocusing their attention on the road). Therefore, exponential decay is used to simulate the natural decay process of risk over time. The distance between the vehicle and the lane line directly affects the collision risk. The farther the deviation (the smaller d), the higher the risk urgency. Linear decay is used to map the physical distance to the risk coefficient, and the two are superimposed to obtain the spatiotemporal attenuation weight.

[0046] For example, a vehicle is driving on a highway. Eye closure is detected at the 30th second and ends at the 32nd second, with a total duration of 2 seconds. Behavior atom A is triggered. At the 35th second, the vehicle deviates from the lane, triggering behavior atom B. At the 40th second, the vehicle speed reaches 85 km / h, triggering behavior atom C. This meets the fatigue-out-of-control behavior chain. Since the current scenario is a high-speed scenario, the basic weight of behavior atom A is recorded as , the basic weight of behavior atom B is recorded as , the basic weight of the behavior atom C is recorded as According to the high-speed scenario rules, the weight coefficient of behavior atom B (lane departure) is 1.2. Since the start time of the behavior chain is 30 seconds, the behavior chain trigger time is 40 seconds, and behavior atom B is triggered at 35 seconds, the time difference from the behavior chain trigger time is 5 seconds, which is the behavior trigger time difference of behavior atom B. , then the spatiotemporal attenuation weight of the behavior atom B is: .

[0047] Step S104: Obtain the spatiotemporal attenuation weight of each behavior atom, combine the spatiotemporal attenuation weight of each behavior atom with the behavior triggering feature value, and obtain the driver's behavior composite risk value. Based on the behavior composite risk value, issue an abnormal warning for the driver's behavior.

[0048] Based on the calculation formula for the spatiotemporal decay weight of any of the aforementioned behavior atoms, the spatiotemporal decay weight of each behavior atom can be derived. Since the concurrent or sequential triggering of multiple behavior atoms can lead to cumulative risk, for example, the combined risk of "eyes closed + lane departure" is much higher than that of a single behavior atom. Therefore, the driver's composite behavior risk value is determined by superimposing the spatiotemporal decay weights of all behavior atoms. The specific method is: based on the spatiotemporal decay weight of each behavior atom, the behavior triggering characteristic values ​​of all behavior atoms are weighted summed to obtain the driver's composite behavior risk value.

[0049] In one embodiment, the calculation formula for the driver's behavior composite risk value is:

[0050] in, represents the composite risk value of the driver's behavior, m represents the number of behavior atoms, represents the spatiotemporal attenuation weight of the i-th behavior atom, It represents the behavior triggering feature value of the i-th behavior atom. If the i-th behavior atom is triggered, the corresponding behavior triggering feature value is 1; if the i-th behavior atom is not triggered, the corresponding behavior triggering feature value is 0.

[0051] Based on the above calculation formula of the behavior composite risk value, the behavior composite risk value of the driver's corresponding behavior chain can be obtained, and then the driver's behavior can be abnormally warned according to the behavior composite risk value. The specific method of abnormal warning is as follows: Obtain the preset first risk warning value and second risk warning value, and the first risk warning value is greater than the second risk warning value. If the behavior composite risk value is greater than the first risk warning value and less than the second risk warning value, a voice prompt is given; if the behavior composite risk value is greater than or equal to the second risk warning value, emergency braking is performed.

[0052] In an implementation manner, the abnormal warning levels are as follows: when R < 1.0, only the log is recorded (such as a short distraction without causing vehicle deviation); when 1.0 < R < 1.5, a voice prompt is given (such as lane departure caused by fatigue but at a low speed); when R ≥ 1.5, emergency braking is performed (such as continuous eyes closed + severe lane departure at high speed). It should be noted that for different scenarios, the risk warning threshold can be adaptively corrected. For example, in rainy days, the visibility is poor and the road surface is slippery, which will cause the driver's reaction time to become longer and the risk to increase, so the high-risk threshold in rainy days can be reduced to R ≥ 1.2.

[0053] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. A driver behavior monitoring and analysis method based on artificial intelligence, characterized in that: The method comprises: While the driver is driving the vehicle, driver characteristic information and vehicle dynamic information are collected and detected. Based on the driver characteristic information and vehicle dynamic information, behavior trigger analysis is performed on at least two behavior atoms to determine the behavior monitoring time. The driver characteristic information includes the duration of continuous eye closure, head deflection angle, and steering wheel grip. The vehicle dynamic parameters include steering wheel angle, vehicle speed, and lane departure distance. Monitor the behavior state changes of each behavior atom within a preset time period after the behavior monitoring moment, perform behavior chain matching on the driver based on the behavior state changes of each behavior atom, and if the match is successful, obtain the real-time lane departure distance, real-time vehicle speed, and the behavior triggering time difference and behavior triggering characteristic value of each behavior atom; For any behavior atom, a basic weight of the behavior atom is obtained according to the driving scenario, and the basic weight of the behavior atom is adjusted according to the real-time lane departure distance, the real-time vehicle speed, and the behavior triggering time difference of the behavior atom to obtain the spatiotemporal attenuation weight of the behavior atom; The spatiotemporal attenuation weight of each behavior atom is obtained, and the composite risk value of the driver's behavior is obtained by combining the spatiotemporal attenuation weight of each behavior atom with the behavior triggering characteristic value. An abnormal warning of the driver's behavior is issued based on the composite risk value of the behavior.

2. The driver behavior monitoring and analysis method based on artificial intelligence according to claim 1 is characterized in that: The behavior atoms include continuous eye closure behavior atoms, lane departure behavior atoms, steering wheel angle high volatility behavior atoms, grip force behavior atoms, and head deviation behavior atoms. Based on the driver characteristic information and vehicle dynamic information, behavior trigger analysis is performed on at least two behavior atoms to determine the behavior monitoring time, including: If it is detected that the driver's continuous eye closure time is greater than or equal to the preset eye closure time, the continuous eye closure behavior atom is triggered; if it is detected that the lane deviation distance is less than or equal to the preset ratio of the lane width and lasts for a first preset time, the lane departure behavior atom is triggered; if it is detected that the standard deviation of the steering wheel angle change rate within a preset time period is greater than a preset change rate standard deviation threshold, the steering wheel angle high volatility behavior atom is triggered; if it is detected that the standard deviation of the steering wheel grip force within a preset time period is less than a preset grip force standard deviation threshold, the grip force behavior atom is triggered; if it is detected that the driver's head deflection angle is greater than a preset angle and lasts for a second preset time period, the head deviation behavior atom is triggered; The time point corresponding to the first triggered behavior atom is determined, and the time point is used as the behavior monitoring moment.

3. The driver behavior monitoring and analysis method based on artificial intelligence according to claim 1 is characterized in that: The driver's behavior chain matching is performed based on the behavior state change of each behavior atom, including: If the continuous closing of eyes behavior atom is triggered continuously within the preset time period, and there is a lane departure behavior atom triggered, and the vehicle speed is greater than the preset speed, then it is determined that the driver meets the fatigue-loss of control behavior chain; If the head deviation behavior atom is triggered continuously within a preset time period and the vehicle's deceleration is greater than the emergency braking deceleration threshold, the driver is determined to have met the distraction-misoperation behavior chain; If the standard deviation of the steering wheel grip force within the preset time period is less than the preset grip force standard deviation threshold, and the standard deviation of the steering wheel angle change rate is greater than the preset change rate standard deviation threshold, it is determined that the driver meets the physiological abnormality-out-of-control behavior chain.

4. The driver behavior monitoring and analysis method based on artificial intelligence according to claim 1 is characterized in that: The method for obtaining the behavior triggering time difference and the behavior triggering characteristic value of each behavior atom includes: For any behavior atom, determine the time point when any behavior atom is triggered for the first time within the preset time length, record it as the initial triggering moment, calculate the time interval between the initial triggering moment and the last moment within the preset time length, record it as the behavior triggering time difference of any behavior atom, if the behavior triggering time difference of any behavior atom is not 0, set the behavior triggering characteristic value of any behavior atom to the first preset value, if the behavior triggering time difference of any behavior atom is 0, set the behavior triggering characteristic value of any behavior atom to the second preset value.

5. The driver behavior monitoring and analysis method based on artificial intelligence according to claim 2 is characterized in that: The driving scenario includes a highway scenario and an urban scenario, and obtaining the basic weight of any behavior atom according to the driving scenario includes: If any of the behavior atoms is a lane departure behavior atom, the basic weight of the behavior atom in the highway scenario is set to a first parameter value, and the basic weight in the urban scenario is set to a second parameter value, and the first parameter value is greater than the second parameter value; If any of the behavior atoms is a head shift behavior atom, the basic weight of the behavior atom in the urban scene is set to the second parameter value, the number of high-risk events in which the behavior atom participated and the total number of high-risk events in the high-speed scene during the historical period are counted, and the ratio of the number of high-risk events in which the behavior atom participated to the total number of high-risk events is calculated as the basic weight of the behavior atom in the high-speed scene; If any of the behavior atoms is not a lane departure behavior atom and a head deviation behavior atom, then the number of high-risk events in which any of the behavior atoms participates and the total number of high-risk events in any scenario during the historical period are counted, and the ratio of the number of high-risk events in which any of the behavior atoms participates to the total number of high-risk events is calculated as the basic weight of any of the behavior atoms in any scenario; wherein, when the vehicle speed is greater than the preset speed, the basic weight of the grip behavior atom in the urban scenario is set to the third parameter value.

6. The driver behavior monitoring and analysis method based on artificial intelligence according to claim 1 is characterized in that: The step of adjusting the basic weight of any behavior atom according to the real-time lane departure distance, the real-time vehicle speed, and the behavior triggering time difference of any behavior atom to obtain the spatiotemporal attenuation weight of any behavior atom includes: Adaptively obtaining a time decay coefficient according to the real-time vehicle speed, and using the inverse of the product of the behavior triggering time difference of any behavior atom and the time decay coefficient as the independent variable of an exponential function with a natural constant as the base to obtain a time adjustment coefficient; Calculate the ratio of the real-time lane departure distance to the lane width, and obtain the difference between the constant 1 and the ratio, which is recorded as the spatial adjustment coefficient; The product of the basic weight of any behavior atom, the time adjustment coefficient and the space adjustment coefficient is used as the spatiotemporal attenuation weight of any behavior atom.

7. The driver behavior monitoring and analysis method based on artificial intelligence according to claim 6 is characterized in that: The adaptively obtaining the time attenuation coefficient according to the real-time vehicle speed includes: Calculate the speed ratio between the real-time vehicle speed and the preset maximum speed, obtain the product between the preset speed sensitivity coefficient and the speed ratio, and use the sum of the constant 1 and the product as the coefficient adjustment value; and use the product of the basic attenuation rate and the coefficient adjustment value as the time attenuation coefficient.

8. The driver behavior monitoring and analysis method based on artificial intelligence according to claim 1 is characterized in that: The composite risk value of the driver's behavior is obtained by combining the spatiotemporal attenuation weight of each behavior atom and the behavior triggering characteristic value, including: Based on the spatiotemporal attenuation weight of each behavior atom, the behavior triggering feature values ​​of all behavior atoms are weighted summed to obtain the driver's behavior composite risk value.

9. The driver behavior monitoring and analysis method based on artificial intelligence according to claim 1 is characterized in that: The abnormal warning of the driver's behavior according to the composite risk value of the behavior includes: Obtain the preset first risk warning value and second risk warning value, and the first risk warning value is greater than the second risk warning value. If the composite risk value of the behavior is greater than the first risk warning value and less than the second risk warning value, a voice prompt is given; if the composite risk value of the behavior is greater than or equal to the second risk warning value, emergency braking is performed.

Images