Learning device and learning method

Abstract

To provide a driving assistance device capable of detecting signs of a driver entering a state of functional decline and warning the driver at an earlier stage so that the driver does not feel inconvenienced.SOLUTION: A driving assistance device includes: a normal following behavior learning unit 016 that learns whether a fluctuation tendency of a following behavior of a driver of a self vehicle following a vehicle ahead is within a normal range based on at least one feature of changes in the state of the self vehicle and changes in the state between the self vehicle and the vehicle ahead, calculated based on predetermined data detected in a time series; a following behavior abnormality determination unit 017 that determines a current following behavior to be abnormal if the fluctuation in the current following behavior differs by more than a certain amount from the normal range; and an alarm control unit 018 that issues an alarm if the current following behavior is determined to be abnormal.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a driving assistance device that prevents or reduces a collision with an obstacle that may occur due to carelessness of a driver of a vehicle when the driver is in a state of mild functional impairment. [Background technology]

[0002] In the technical field, an invention is known relating to a driver condition detection device that detects abnormalities in a driver who has fallen into a state of mild functional impairment using a forward vehicle detection sensor and an acceleration / deceleration sensor that detects acceleration / deceleration (see Patent Document 1 below).

[0003] For example, Patent Document 1 discloses a driver condition detection device that detects abnormalities in the driver, and includes a forward vehicle detection sensor that detects vehicles traveling in front of the host vehicle or to the side of the host vehicle and in front of the host vehicle, an acceleration / deceleration sensor that detects the acceleration / deceleration of the host vehicle, an acceleration / deceleration calculation unit that calculates, based on an acceleration / deceleration model, an appropriate acceleration / deceleration for causing the host vehicle to travel so as to follow the preceding vehicle detected by the forward vehicle detection sensor, and an abnormality determination unit that compares the acceleration / deceleration calculated by the acceleration / deceleration calculation unit with the actual acceleration / deceleration of the host vehicle detected by the acceleration / deceleration sensor to determine whether or not there is an abnormality in the driver.The abnormality determination unit is characterized in that, when the forward vehicle detection sensor detects another vehicle that may be coming between the host vehicle and the preceding vehicle that is being followed, the degree of agreement between the acceleration / deceleration calculated by the acceleration / deceleration calculation unit and the actual acceleration / deceleration of the host vehicle is higher than a predetermined threshold. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Patent Publication No. 2021-160626 Summary of the Invention [Problem to be solved by the invention]

[0005] However, in Patent Document 1, the impaired function state is detected when the difference between the driver's acceleration / deceleration rate and the appropriate acceleration / deceleration rate for following the vehicle ahead exceeds a predetermined threshold. Therefore, the impaired function state can only be detected when there is a delay in the driver's operation relative to the vehicle ahead. Therefore, if the deceleration rate of the vehicle ahead is strong, even if an alarm is issued after the impaired function state is detected, there may not be enough time for the driver's function to return to normal. Furthermore, while it is possible to speed up the detection of the impaired function state by reducing the threshold, setting it too early may result in false detection of the impaired function state, which may be annoying to the driver.

[0006] Therefore, in view of the above circumstances, the present invention aims to provide a driving assistance device that can detect signs that a driver is entering a state of functional decline and warn the driver at an earlier stage so that the driver does not feel inconvenienced. [Means for solving the problem]

[0007] In order to achieve the above object, the driving assistance device of the present invention is characterized by comprising: a leading vehicle recognition unit that detects leading vehicles traveling in front of the vehicle; a following behavior normality learning unit that learns whether the tendency for fluctuations in the following behavior of the driver of the vehicle to follow the leading vehicle is within a normal range from at least one characteristic amount of changes in the state of the vehicle calculated based on predetermined data detected in a time series and changes in the state between the vehicle and the leading vehicle; a following behavior abnormality determination unit that determines the current following behavior to be abnormal if the fluctuations in the current following behavior differ by more than a certain amount from the normal range; and an alarm control unit that issues an alarm if the current following behavior is determined to be abnormal.By providing a driving assistance device, it is possible to detect signs that the driver is entering a state of functional decline and warn the driver at an earlier stage. [Effects of the Invention]

[0008] According to the present invention, by detecting a sign that a driver is entering a state of impaired function, it is possible to warn the driver at an earlier stage and prevent or mitigate a collision with an obstacle.

[0009] Problems, configurations, and effects other than those described above will become apparent from the following description of the embodiments. [Brief explanation of the drawings]

[0010] [Figure 1] 1 is a configuration diagram of a driving assistance device according to an embodiment of the present invention; [Figure 2] 3 is a flowchart of a driving assistance device according to an embodiment of the present invention. [Figure 3] 10 shows a first scene of suppressing a determination of an abnormality in a following operation of the driving assistance device according to the embodiment of the present invention. [Figure 4] 10 shows a second scene in which the driving assistance device according to the embodiment of the present invention suppresses the determination of an abnormality in the following operation. [Figure 5] 10 shows a third scene in which the driving assistance device according to the embodiment of the present invention suppresses the determination of an abnormality in the following operation. [Figure 6] 10 shows an example of fluctuation tendency of the following operation of the driving assistance device according to one embodiment of the present invention. [Figure 7] 3 is an example of an amplitude spectrum of the driving assistance device according to the embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0011] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS A driving assistance device according to an embodiment of the present invention will now be described with reference to the accompanying drawings.

[0012] <Example 1: Example in which learning results are parameters> FIG. 1 is a configuration diagram of a driving assistance device according to an embodiment of the present invention.

[0013] [Configuration explanation] The driving assistance device 010 of this embodiment is an electronic control unit (ECU) mounted on a vehicle (host vehicle) 001, such as a gasoline vehicle, a diesel vehicle, a natural gas vehicle, a hybrid vehicle, an electric vehicle, a fuel cell vehicle, or a hydrogen engine vehicle. Although not shown, the driving assistance device 010 is configured by one or more microcontrollers including, for example, an input / output unit, a central processing unit (CPU), memory (including both non-volatile memory and volatile memory), and a timer.

[0014] (Vehicle 001) The host vehicle 001 is equipped with, for example, a host vehicle sensor 002, an external sensor 003, a car navigation system (CNS) 008, an audio output device 004, an image display device 005, an acceleration device 006, and a deceleration device 007. Although not shown, the host vehicle 001 also includes a drive system, a steering system, a braking system, and a control system for running, turning, decelerating, and stopping the host vehicle 001.

[0015] (Vehicle sensor 002) The host vehicle sensor 002 includes various sensors that detect the state of the host vehicle, such as a wheel speed sensor, an acceleration sensor, a shift position sensor, a gyro sensor, a steering angle sensor, and a turn signal sensor, and detects the host vehicle state, including the speed, acceleration, transmission shift position, yaw rate, steering angle, and turn signal operation state of the host vehicle 001, as well as any abnormalities of the host vehicle 001, and outputs the detected results to the driving assistance device 010. Abnormalities of the host vehicle 001 detected by the host vehicle sensor 002 include, for example, abnormalities in tire pressure, remaining fuel, engine, ABS (Anti-Lock Brake System), airbags, brakes, oil pressure, battery, water temperature, etc.

[0016] (External Sensor 003) The external sensor 003 may include, for example, a millimeter-wave radar that uses reflected waves of radio waves such as millimeter waves, a monocular camera, a stereo camera, a LiDAR (Light Detection and Ranging, Laser Imaging Detection and Ranging) that measures scattered light from pulsed laser irradiation to determine the distance to an object, an ultrasonic sensor that uses reflected waves of ultrasonic waves, a road-to-vehicle communication device, an inter-vehicle communication device, an illuminance sensor, a raindrop sensor, a humidity sensor, etc. The external sensor 003 detects objects around the host vehicle 001, such as roads, lane lines, signs, traffic signals, vehicles, pedestrians, and obstacles, as well as the surrounding environment, such as illuminance, rainfall, humidity, and the visibility of obstacles, and outputs the detected information to the driving assistance device 010. Using FIG. 3 as an example, the millimeter-wave radar, monocular camera, stereo camera, LiDAR, etc. of the external sensor 003 detects vehicles 402 and 403 within a detection range 401 of the host vehicle 400 and outputs the detected information to the driving assistance device 010. Note that the detection range 401 is an example.

[0017] (CNS008) The CNS 008 includes, for example, a map information storage device, a route calculation device, a vehicle-to-vehicle communication device, a road-to-vehicle communication device, and a Global Navigation Satellite System (GNSS) receiver. The CNS 008 also includes, for example, an input device for the driver of the vehicle 001 to input a destination. The CNS 008 outputs, to the driving assistance device 010, for example, point information on a map based on the position information of the vehicle 001, route information from the current position of the vehicle 001 to the destination, intersection position information that appears between the current position and the destination, curve information such as the radius of curvature, gradient information, stop line information, lane width information, traffic signal information, and the like.

[0018] (Audio Output Device 004) The audio output device 004 is, for example, a speaker provided in the cabin of the vehicle 001 , and outputs warning sounds and audio guidance based on a control signal input from the driving assistance device 010 .

[0019] (Image display device 005) The image display device 005 is, for example, a liquid crystal display device, an organic EL display device, or a head-up display, and displays an image based on a control signal input from the driving assistance device 010. The image display device 005 may also include an input device such as a touch panel or operation buttons. The driver of the vehicle 001 can input information such as a destination to the CNS 008 via the input device of the image display device 005. The driver may also be able to input a determination result regarding whether the warning sound or warning display output to the audio output device 004 or the image display device 005 is correct via the input device of the image display device 005. In addition, the driver may be able to select which parameter to use from multiple parameters used for processing in the driving assistance device 010.

[0020] (Accelerator 006) The acceleration device 006 is, for example, an engine or a motor, and accelerates the host vehicle 001 based on a control signal input from the driving support device 010. In addition, based on a request for suppressing acceleration input from the driving support device 010, the acceleration device 006 has a function of not accelerating the host vehicle 001 even if the driver steps on the accelerator.

[0021] (Decelerator 007) The deceleration device 007 is, for example, a brake, and decelerates the host vehicle 001 based on a control signal input from the driving support device 010 .

[0022] (Driving assistance device 010) The driving assistance device 010 of this embodiment is mounted on the vehicle 001 and functions as a driving assistance device that detects signs of a functional decline in the driver of the vehicle 001 and assists the driver in restoring normal functionality. The driving assistance device 010 is a driving assistance device that includes a surrounding environment recognition unit 013, a leading vehicle recognition unit 014, a following operation abnormality determination suppression unit 015, a following operation normality learning unit 016, a following operation abnormality determination unit 017, an alarm control unit 018, and a speed control unit 019.

[0023] (Surrounding Environment Awareness Department 013) The surrounding environment recognition unit 013 detects object information and road information around the vehicle. For example, the object information refers to objects such as vehicles (including four-wheeled vehicles, two-wheeled vehicles, and bicycles) traveling in lanes adjacent to the vehicle, pedestrians, and obstacles. Furthermore, the road information refers to information such as intersection position information, curve information including curvature radius, stop line information, lane width information, traffic signal position and traffic signal status, etc.

[0024] The surrounding environment recognition unit 013 outputs the recognition results, such as the position, speed, acceleration / deceleration of surrounding objects, and the position of surrounding roads, to the following operation abnormality determination suppression unit 015.

[0025] (Vehicle in front recognition unit 014) The vehicle-in-front recognition unit 014 detects a vehicle in front that is traveling ahead of the host vehicle. Using FIG. 3 as an example, a vehicle 402 traveling ahead of the host vehicle 400 is an example of a vehicle in front. "Ahead" refers to an area in the direction of travel within the lane of the host vehicle. Note that a vehicle in front includes four-wheeled vehicles, two-wheeled vehicles, and bicycles, and refers to an object within a predetermined distance of a detection range 401 in front of the host vehicle.

[0026] The vehicle-in-front recognition unit 014 outputs the recognition results such as the position, speed, and acceleration / deceleration of the vehicle in front to the following operation abnormality determination suppression unit 015 and the following operation normality learning unit 016 .

[0027] (Follow-up operation abnormality determination suppression unit 015) The following operation abnormality determination suppression unit 015 determines whether to suppress the abnormality determination by the following operation abnormality determination unit 017. This determination can exclude from processing situations in which the following operation abnormality determination unit 017 may make an erroneous abnormality determination. In other words, it determines (predicts) whether the current fluctuation in the following operation is normal or a transient state in which an abnormality cannot be determined. The fluctuation in the following operation refers to the relative movement between the host vehicle and the vehicle ahead when the driver of the host vehicle manually follows the vehicle ahead. The relative movement refers to a change in the state of the host vehicle detected over time by the host vehicle sensor 002 (calculated based on predetermined data) or a change in the state between the host vehicle and the vehicle ahead detected by the external sensor 003. Specifically, the change in the state of the host vehicle includes a change in acceleration / deceleration and a change in the host vehicle speed. The change in the state between the host vehicle and the vehicle ahead includes a change in the relative speed, a change in the inter-vehicle distance, a change in the relative acceleration, etc.

[0028] Generally, when a driver manually follows a vehicle in front, the driver follows the vehicle so that the distance and relative speed between the vehicle and the vehicle in front approach the values ​​the driver desires. It is known that fluctuations in the following behavior occur during this process. For example, fluctuations in the following behavior can be understood as changes in relative speed as shown in Figure 6. In Figure 6, normal fluctuations in the following behavior refer to changes in relative speed 501 fluctuating within a range of ±5 km / h around 0 km / h (5 km / h is an example). Abnormal fluctuations in the following behavior refer to cases in which changes in relative speed 502 exceed the range of ±5 km / h, and the relative speed fluctuates less frequently than when the following behavior is normal.

[0029] A situation in which the following operation abnormality determination unit 017 may make an erroneous determination of an abnormality refers to, for example, a situation in which the vehicle ahead suddenly accelerates or decelerates. This is because the change in inter-vehicle distance and relative speed varies depending on the acceleration or deceleration of the vehicle ahead, and it is expected that the driver will perform a following operation that differs from normal. Therefore, since it is difficult to determine whether the following operation is normal or abnormal, abnormal determination is suppressed. Sudden acceleration or deceleration can be determined when the deceleration of the vehicle ahead, as detected by the vehicle ahead recognition unit 014, is equal to or greater than a predetermined threshold, or when the acceleration is equal to or greater than a predetermined threshold.

[0030] Furthermore, if it is possible to determine whether the blinker of the vehicle ahead is on from information from the monocular camera or stereo camera of the external sensor 003, it is possible to predict that the vehicle ahead will depart from the vehicle's own lane (driving lane) in the future. Therefore, the driver may intentionally accelerate or decelerate the vehicle, changing the inter-vehicle distance or relative speed, making it impossible to uniquely predict the driver's movements. Therefore, since it is difficult to determine whether the following operation is normal or abnormal, abnormality determination is suppressed.

[0031] Furthermore, if there is a sharp curve or a steep gradient ahead of the vehicle in front, there is a high possibility that the vehicle in front will slow down. However, the driver's acceleration / deceleration behavior changes depending on whether or not the driver is aware of the sharp curve or steep gradient ahead, so the driver's movements cannot be uniquely predicted. Therefore, since it is difficult to determine whether the following behavior is normal or abnormal, abnormality determination is suppressed. Note that a sharp curve can be determined by whether the radius of curvature of the curve from the monocular camera or stereo camera of the external sensor 003 or the CNS 008 falls within a predetermined threshold range. A steep gradient can be determined by whether the road gradient from the CNS 008 falls within a predetermined threshold range.

[0032] Furthermore, if the presence of a stop line ahead of the leading vehicle can be determined from the stop line information from CNS008, there is a high possibility that the leading vehicle will slow down. However, the driver's deceleration behavior changes depending on whether or not the driver is aware of the stop line ahead, so the driver's behavior cannot be uniquely predicted. Therefore, since it is difficult to determine whether the following behavior is normal or abnormal, abnormality judgments are suppressed.

[0033] Furthermore, if it can be determined from information from the monocular camera or stereo camera of the external sensor 003 and lane width information from the CNS 008 that the lane width ahead of the leading vehicle is decreasing, there is a high possibility that the leading vehicle will slow down. However, whether the leading vehicle actually slows down depends solely on the awareness of the driver of the leading vehicle. Therefore, since it is difficult to determine whether the following operation is normal or abnormal, abnormality determination is suppressed. Note that the decrease in lane width can be determined from the amount of change in lane width.

[0034] Furthermore, if the monocular camera or stereo camera of the external sensor 003 can determine that there is a red light ahead of the leading vehicle, there is a high possibility that the leading vehicle will slow down. However, the driver's deceleration behavior changes depending on whether or not the driver can recognize the color of the traffic light ahead, so the driver's behavior cannot be uniquely predicted. Therefore, since it is difficult to determine whether the following behavior is normal or abnormal, abnormality determination is suppressed.

[0035] 4, if it is possible to determine from information from the monocular camera or stereo camera of the external sensor 003 that a pedestrian 405 is walking near the vehicle's own lane, the driver may decelerate regardless of the movement of the preceding vehicle 406. This may cause the driver to change their approach to vehicle-following operations, and the top priority may be given to avoiding a collision with the pedestrian 405. Therefore, since it is difficult to determine whether the vehicle-following operation is normal or abnormal, the abnormality determination is suppressed.

[0036] Furthermore, as shown in FIG. 3, if it is possible to determine from information from the monocular camera or stereo camera of the external sensor 003 that an adjacent vehicle 403 is about to cut in front of the vehicle, the driver may decelerate to allow the adjacent vehicle 403 to cut in, regardless of the movement of the preceding vehicle 402, or the driver may accelerate to prevent the adjacent vehicle 403 from cutting in. Therefore, the driver's movement cannot be uniquely predicted, making it difficult to determine whether the following behavior is normal or abnormal, and therefore abnormality determination is suppressed. Note that cutting in can be determined by whether the relative lateral speed of the adjacent vehicle 403 is approaching the vehicle at or above a predetermined threshold. Alternatively, it may be determined by whether the relative lateral position of the adjacent vehicle 403 is within a predetermined threshold range after a predetermined time.

[0037] Furthermore, if it can be determined from the acceleration / deceleration information of the acceleration sensor of the host vehicle sensor 002 that the host vehicle has suddenly accelerated or decelerated, there is a high possibility that an event has occurred that the driver needs to prioritize over following the vehicle ahead. Therefore, since it is difficult to determine whether the following operation is normal or abnormal, abnormality determination is suppressed. Furthermore, sudden acceleration / deceleration can be determined by whether the acceleration / deceleration is within a predetermined threshold range.

[0038] Furthermore, if it can be determined from the turn signal of the host vehicle sensor 002 that the driver has operated the turn signal, it can be predicted that the driver will stop following the vehicle ahead and move to another lane. Therefore, there will no longer be an object ahead of the host vehicle. Therefore, the abnormality determination is suppressed.

[0039] Furthermore, as shown in FIG. 5, in a situation where there is a road 407 merging into the driving lane of the host vehicle, if a merging vehicle 404 is present on the road 407, the driver may decelerate the host vehicle, thinking that the merging vehicle 404 may merge into the driving lane of the host vehicle, regardless of the movement of the preceding vehicle 408. This may temporarily change the inter-vehicle distance and relative speed with the preceding vehicle 408, which may lead to an erroneous determination that the following operation is abnormal. Therefore, to prevent an erroneous determination of an abnormality in the following operation, the abnormality determination is suppressed. Note that a merging situation can be determined when the relative lateral position of the merging vehicle 404 is within a predetermined threshold. Furthermore, whether a merging road exists ahead of the host vehicle may be added to the conditions for determining a merging situation from the CNS 008.

[0040] Furthermore, if the reliability of the recognition of the vehicle ahead has decreased, the accuracy of the inter-vehicle distance and relative speed from the vehicle ahead from the external sensor 003 is likely to have decreased, and therefore abnormality determination is suppressed. The reliability of the recognition of the vehicle ahead can be determined by determining whether the state in which detection of the vehicle ahead continues from the start of detection is within a predetermined time. Alternatively, it may be determined from information related to the recognition reliability from the external sensor 003. Alternatively, it may be determined by determining whether there is a sudden change in the current inter-vehicle distance or relative speed value that exceeds a predetermined threshold range compared to the past inter-vehicle distance or relative speed value.

[0041] That is, the follow-up operation abnormality determination suppression unit 015 predicts a transient state in which it is not possible to determine whether the current fluctuation in the follow-up operation is normal or abnormal, and suppresses the abnormality determination by the follow-up operation abnormality determination unit 017. When the preceding vehicle suddenly accelerates or decelerates, The blinker of the car in front is activated, There is a sharp curve ahead of the vehicle in front, A situation where there is a steep slope ahead of the leading vehicle, There is a stop line ahead of the vehicle in front, There is a red light ahead of the vehicle in front, The lane width in front of the vehicle in front is narrowing, A state in which a pedestrian is walking near the vehicle's lane, A situation in which it is possible to predict that an adjacent vehicle will cut in front of your vehicle, A state in which sudden acceleration or deceleration of the vehicle occurs, The driver of the vehicle operates the turn signal. A state in which there is an object whose advance is predictable in the vehicle's lane, and a state in which the reliability of recognition of the vehicle in front is reduced.

[0042] (Follow-up operation normal learning unit 016) The normal following behavior learning unit 016 learns the normal range of the following behavior when the driver of the host vehicle is following a vehicle in front. Then, based on the learned normal range of the following behavior, if the current following behavior is within the normal range, it is determined that the current following behavior is normal. In other words, the normal following behavior learning unit 016 learns whether the tendency of fluctuation in the driver's following behavior of the vehicle in front is within the normal range. Note that the normal range refers to the following behavior in a state where the driver of the host vehicle is paying attention to driving and is following the vehicle in front.

[0043] In this embodiment, it is assumed that the normal range of the following operation is stored in the non-volatile memory of the driving assistance device 010 as a parameter based on a predetermined threshold learned in advance, and the parameter is read and used when the driving assistance device 010 is started. Then, the current following operation of the host vehicle is calculated from changes in the state of the host vehicle and changes in the state between the host vehicle and the vehicle ahead, and whether the following operation is normal is determined by comparing it with the learned parameters in the normal range.

[0044] To calculate the current tracking operation, for example, the frequency, amplitude, and bandwidth that represent the tendency of change in relative speed are calculated. The calculation method is as follows.

[0045] First, since the relative speed from the external sensor 003 contains noise, the noise is removed by filtering, such as by moving average, as a preprocessing step. The filtering method may be switched to another method (such as a band-pass filter or a high-pass filter) depending on the specifications of the input information from the external sensor 003. Furthermore, when the host vehicle speed is slower than when the host vehicle speed is fast, it is easier to start or change lanes, and the probability of sudden acceleration or deceleration of the host vehicle is considered to be higher. Therefore, when the acceleration or deceleration estimated according to the host vehicle speed is equal to or greater than a predetermined threshold, it may be determined that this is a temporary deviation from the following operation, and filtering may be performed to remove the change in relative speed at that time. Filtering may be performed to remove changes in relative speed due to temporary deviations from the following operation, depending not only on the host vehicle but also on the speed and acceleration / deceleration of the vehicle ahead.

[0046] Next, to calculate the frequency, amplitude, and bandwidth representing the change in relative speed, it is necessary to specify the range of data to be analyzed. In this embodiment, time-series data is acquired, extracting relative speed values ​​from a predetermined time ago (past) to the present. Since there are more disturbance factors that interfere with the host vehicle's tracking on ordinary roads than on expressways, the duration of the tracking operation may vary depending on the host vehicle's speed range. Disturbance factors refer to factors that temporarily prevent the host vehicle from continuing tracking, such as a vehicle cutting into the host vehicle's lane or a pedestrian crossing the road. For example, when the host vehicle is traveling at a slow speed, it is often traveling on ordinary roads, where there are more disturbance factors that interfere with the host vehicle's tracking operation, resulting in a shorter duration of tracking operation than on expressways. Taking advantage of this characteristic, the predetermined time may be varied depending on the host vehicle's speed. Alternatively, the predetermined time may be varied depending on the road type, using information from CNS008 indicating whether the vehicle is traveling on an ordinary road or an expressway. It is also desirable to set the predetermined time to a relatively long period, from several tens of seconds to several minutes, so that the tracking operation can be captured.

[0047] Then, a fast Fourier transform (FFT) or the like is performed based on the obtained time-series data of the relative velocity, thereby performing frequency conversion and calculating an amplitude spectrum. For example, assume an amplitude spectrum 701 as shown in FIG. 7. The amplitude spectrum 701 has a frequency F1 at which an amplitude P1 has a peak, and has a bandwidth 703. If the amplitude P1 is within a predetermined threshold, the amplitude is considered to be within a normal range. If the frequency F1 is within a predetermined threshold, the frequency is considered to be within a normal range. If the bandwidth 703 is within a predetermined threshold, the bandwidth is considered to be within a normal range. Note that if all of these conditions are met, the following behavior by the driver is considered to be within a normal range. However, depending on the time-series data, a peak value may not necessarily occur only at frequency F1 as shown in FIG. 7. Therefore, if multiple amplitudes of the same level and multiple different frequencies appear in the frequency spectrum, the following behavior may not be determined to be normal.

[0048] Finally, if the follow-up operation remains within the normal range for a predetermined period of time, it is determined that the follow-up operation is normal.

[0049] Here, relative speed is used as an example of a change in the relative relationship between the vehicle in front and the vehicle itself, but a similar fast Fourier transform calculation can also be performed using the inter-vehicle distance or THW (Time Headway), which is the inter-vehicle distance divided by the vehicle's speed.

[0050] That is, the normal following operation learning unit 016 calculates one or more of frequency, amplitude, and bandwidth by frequency converting the change in the relative relationship between the vehicle in front and the vehicle itself (relative speed, vehicle-to-vehicle distance, THW, etc.), and learns that if one or more of the calculated frequency, amplitude, and bandwidth are within a predetermined threshold range, they are normal (included in the normal range).

[0051] It is generally known that when a driver follows a vehicle ahead, the THW averages approximately 2 seconds. Therefore, instead of using frequency conversion, a normal range of 2 seconds plus or minus a predetermined threshold may be defined, and the following behavior may be determined to be normal if the average THW falls within the predetermined threshold. The predetermined threshold can be determined theoretically by extracting a portion of the time-series data where the following behavior is within the normal range and using the standard deviation calculated and analyzed for the THW. While the average THW is approximately 2 seconds when the road is not congested, it can be expected to be shorter than 2 seconds when the road is congested. This is due to the driver's intention to close the following distance to prevent the adjacent vehicle from cutting in front of the vehicle during congested roads. Therefore, it is recommended to set the average THW parameter to a value shorter than 2 seconds, such as approximately 1.5 seconds, when the road is congested. These average THW values ​​vary depending on the individual, so 2 and 1.5 seconds are merely examples.

[0052] The average value of the change in the relative relationship between the vehicle in front and the own vehicle may be calculated based on the relative speed or the distance between the vehicles.

[0053] That is, the normal following operation learning unit 016 may learn that the change in the relative relationship between the vehicle in front and the own vehicle (relative speed, inter-vehicle distance, THW, etc.) is normal (included in the normal range) when the average value is within a predetermined threshold value.

[0054] Furthermore, the predetermined threshold parameters representing the normal range of the following behavior may be acquired from the cloud via the Internet as learned parameters for drivers not only of the driver of the vehicle but also of drivers nationwide, and stored in a non-volatile memory in the driving assistance device 010 of the vehicle. That is, the normal following behavior learning unit 016 may acquire learned parameters (representing the normal range) (for each driver) from the cloud via the Internet. The learned parameters acquired from the cloud may not be a single parameter, but may be selected from multiple parameters that match the driver's conditions. For example, the determination may be based on scored information on the driver's driving behavior, such as the driver's age, driving history, number of accidents, and whether or not the driver normally accelerates or decelerates suddenly. Furthermore, the driver may be able to select learned parameters that suit him or her using a touch panel or operation buttons on the image display device 005.

[0055] (Follow-up operation abnormality determination unit 017) The following operation abnormality determination unit 017 determines whether the current following operation of the driver of the host vehicle when following a leading vehicle is abnormal based on the normal range and normal determination of the following operation normal learning unit 016. The following operation abnormality determination unit 017 determines that the driver's current following operation is abnormal if the state changes from a state determined to be normal by the following operation normal learning unit 016 to a state deviating from the normal range. A state deviating from the normal range refers to a state in which the fluctuation in the current following operation differs from the normal range by a certain amount or more. In other words, the following operation abnormality determination unit 017 determines that the current following operation is abnormal if the fluctuation in the current following operation differs from the normal range by a certain amount or more.

[0056] Whether or not the condition is outside the normal range is determined as follows.

[0057] When the amplitude P2 of the amplitude spectrum 702 in FIG. 7 is greater than a predetermined threshold, it is determined that the amplitude deviates from the normal range. When the frequency F2 is outside the range of the predetermined threshold, it is determined that the frequency deviates from the normal range. When the bandwidth is greater than the predetermined threshold, it is determined that the bandwidth deviates from the normal range. When any of these conditions is met, it is determined that the driver's tracking behavior deviates from the normal range. When the tracking behavior continues to deviate from the normal range for a predetermined period of time, it is determined that the tracking behavior is abnormal. In other words, when the fluctuation of the current tracking behavior deviates from the normal range by a certain amount or more for a predetermined period of time, it is determined that the tracking behavior is abnormal. Note that these normal ranges may have a margin in the predetermined threshold to perform hysteresis processing to prevent frequent alternation between normal and abnormal.

[0058] Furthermore, if the vehicle has been following a leading vehicle at the same vehicle speed for a long period of time, the following operation is likely to become abnormal due to driver fatigue, etc. Therefore, a condition may be added to determine whether the following operation is abnormal if the speed change from the vehicle speed at the time when the following operation was determined to be normal continues to be within a predetermined range for a predetermined period of time.

[0059] (Alarm control unit 018) When the tracking operation abnormality determination unit 017 determines that there is an abnormality in the current tracking operation, the warning control unit 018 requests (instructs) the audio output device 004 to sound an alarm. The alarm may be a beep or a voice guidance notifying the driver of a functional decline. The warning control unit 018 also requests (instructs) the image display device 005 to display an alarm.

[0060] The alarm control unit 018 may be implemented to request sounding of an alarm or display of an alarm when an abnormality in the tracking operation is confirmed and the confirmation continues for a predetermined period of time, in order to prevent false alarms.

[0061] If an alarm is sounded and a display is output, but the driver is not actually in a state of impaired function, the driver operates the image display device 005 to correct the error in the tracking operation abnormality determination unit 017. For example, a method can be considered in which an operation button that can select whether the alarm is true or false is provided on the alarm display, and the driver operates it to select whether the alarm is true or false.

[0062] If the driver selects that the warning is false, the normal following operation learning unit 016 changes the learned normal range of the following operation to "unlearned" because it is a false alarm. After changing to "unlearned," the driver may again operate the image display device 005 to select a parameter different from the parameter representing the normal range in the normal following operation learning unit 016 used when a false alarm was issued. The different parameter may be, for example, a parameter that allows the driver to specify the tendency of fluctuation in the following operation from large, medium, or small fluctuation in the following operation. When the driver resets the parameter, the normal following operation learning unit 016 corrects the normal range to "learned," and the normal following operation abnormality determination unit 017 is again ready to determine an abnormality. That is, the normal following operation learning unit 016 corrects the learned normal range based on the driver's response.

[0063] Once an alarm has been initiated, the alarm control unit 018 cancels the alarm when a predetermined time has elapsed since the condition for initiating the alarm is no longer satisfied. Alternatively, the alarm control unit 018 cancels the alarm when the driver's intention to drive can be confirmed after the condition for initiating the alarm is no longer satisfied. Once an alarm has been initiated (in other words, while the alarm control unit 018 is instructing an alarm), the driver may still be in a state of impaired function, so measures to reduce erroneous alarm cancellation are necessary. For example, after an abnormality in the following operation is confirmed and an alarm is initiated, the alarm is canceled when a predetermined time has elapsed since the following operation abnormality determination unit 017 determined that the following operation is no longer abnormal (in other words, when a predetermined time has elapsed since the abnormality in the following operation abnormality determination unit 017 was resolved). Furthermore, if a leading vehicle no longer exists during an alarm request, the alarm is canceled when a predetermined time has elapsed since the leading vehicle no longer exists. Furthermore, after an alarm is initiated, the alarm is canceled when a predetermined time has elapsed since the suppression by the following operation abnormality determination suppression unit 015 began. Furthermore, after the warning is initiated, the warning is cancelled when a predetermined time has elapsed since the learning in the following operation normality learning unit 016 was reset. Furthermore, after the warning is initiated, the warning is cancelled when a predetermined time has elapsed since the driver operated the accelerator, brake, blinker, or steering in a state in which the following operation abnormality determination unit 017 has determined that the following operation is normal (in other words, in a state in which the abnormality in the following operation abnormality determination unit 017 has been resolved). The conditions for cancellation may be applied to the speed control in the speed control unit 019 below.

[0064] That is, in a state where the alarm control unit 018 is instructing an alarm, When the abnormality in the follow-up operation abnormality determination unit 017 is resolved, or The time when suppression by the follow-up operation abnormality determination suppression unit 015 starts, or When the learning in the normal following operation learning unit 016 is reset, or When the abnormality detected by the following operation abnormality determination unit 017 is resolved, the warning control unit 018 cancels the warning when a predetermined time has elapsed since the driver started operating the accelerator, brake, turn signal, or steering.

[0065] (Speed ​​control unit 019) If the driver does not change his driving behavior after the warning control unit 018 outputs a warning, the speed control unit 019 requests the acceleration device 006 to suppress acceleration. When the speed control unit 019 requests acceleration suppression, the host vehicle 001 does not accelerate even if the driver presses the accelerator pedal. This reduces the possibility of a collision with a vehicle ahead if the driver accidentally presses the accelerator pedal while in a state of impaired function.

[0066] Alternatively, the speed control unit 019 requests the deceleration device 007 to decelerate. When the request for deceleration is made, the host vehicle 001 automatically starts decelerating. This reduces the possibility of a collision with a vehicle ahead when the driver is in a functionally impaired state and brakes late. The requested deceleration should be calculated based on a constant acceleration linear motion model so that THW does not become equal to or less than a predetermined value.

[0067] Whether the driver continues to change their driving behavior is determined based on whether a predetermined time has elapsed since the warning was issued, or whether the risk of collision with the vehicle ahead has reached a predetermined threshold or higher. The risk of collision can be measured using TTC (Time to Collision) or THW as an index.

[0068] [Flowchart explanation] 2 is a flowchart showing an example of a driving assistance routine executed by the driving assistance device 010. This flowchart is repeatedly executed at a predetermined interval by a CPU included in an ECU that implements the driving assistance device 010.

[0069] When the driving assistance routine is started, in step 300, the vehicle-in-front recognition unit 014 determines whether or not a vehicle in front is present. If it is determined that no vehicle in front is present, no abnormality determination is made in the follow-up operation, and the process proceeds to step 316. Note that it is also advisable to proceed to step 316 if an abnormality has occurred in the host vehicle 001. If it is determined that a vehicle in front is present, in step 302, the follow-up operation abnormality determination suppression unit 015 determines whether or not a condition for suppressing an abnormality determination in the follow-up operation is met. If it is determined that an abnormality determination in the follow-up operation should be suppressed, no abnormality determination in the follow-up operation is made, and the process proceeds to step 316.

[0070] If it is determined that the abnormality determination of the following operation should not be suppressed, the following operation normal learning unit 016 determines whether the normal range has already been learned in step 305. If the normal range has not been learned, the normal range of the following operation is learned in step 312.

[0071] Then, in step 313, the learned result of the normal range is recorded so that the learned normal range can be used the next time the driving assistance device 010 is powered on even if the driving assistance device 010 is powered off. The learned result is stored in a nonvolatile memory of the driving assistance device 010. The stored learned result is read from the nonvolatile memory the next time the driving assistance device 010 is powered on and the vehicle is started.

[0072] Alternatively, the learning results may be stored in a cloud server via the Internet. The stored learning results are read from the cloud server the next time the driving assistance device 010 is powered on and the subject vehicle is started, and are stored in a non-volatile memory within the driving assistance device 010.

[0073] That is, the normal following operation learning unit 016 stores the learned normal range for each driver (in non-volatile memory or a cloud server), and reads out (from the non-volatile memory or the cloud server) the stored normal range for the corresponding driver the next time the vehicle is started when the driving assistance device 010 is powered on.

[0074] After recording in step 313, proceed to step 316.

[0075] If the normal range has been learned, the following operation normal learning unit 016 evaluates the current following operation in step 306. The following operation calculated for the current following operation using a method such as the fast Fourier transform (FFT) described above is compared with the learned normal range to determine whether the current following operation is normal (whether it is included in the normal range).

[0076] Next, in step 307, the tracking operation abnormality determination unit 017 determines whether the previous tracking operation was normal. If the previous tracking operation was not normal, the process proceeds to step 316. If the previous tracking operation was normal, the current tracking operation is compared with the learned normal range in step 308, and it is determined whether the fluctuation of the current tracking operation differs from the learned normal range by a certain amount or more. If there is no difference by a certain amount or more, the process proceeds to step 316. If there is a difference by a certain amount or more, the current tracking operation is determined to be abnormal, and the process proceeds to step 309.

[0077] In step 309, the alarm control unit 018 issues an instruction to activate an alarm. In step 316, the alarm control unit 018 determines whether the alarm cancellation conditions are met. If the alarm cancellation conditions are not met, the routine ends without doing anything. If the alarm cancellation conditions are met, the alarm is cancelled in step 317. Then, the routine ends.

[0078] Once the routine is completed, the CPU will execute from step 300 in the next cycle.

[0079] [Effect description] As described above, by detecting an abnormality in the driver's following behavior and issuing an alarm or suppressing acceleration or decelerating, the possibility of a collision can be reduced.

[0080] <Example 2: Learning Example> The configuration diagram of the driving support device according to the embodiment of the present invention is the same as that of the first embodiment, and only the cases where different processes are performed in each configuration of FIG. 1 will be described below.

[0081] [Configuration explanation] The host vehicle sensor 002 includes, in addition to the contents of the first embodiment, for example, a driver monitor which is a camera that captures the interior of the vehicle. The driver monitor detects whether the driver is in a state of impaired function from, for example, the driver's line of sight, facial direction, facial expression, etc., and outputs the result to the driving assistance device 010. The impaired function state refers to the driver's absent-minded driving or drowsiness.

[0082] The driving assistance device 010 has, for example, a non-volatile memory for storing training data in addition to the contents of Example 1. Training data refers to learning data that serves as a correct answer for use in supervised learning in machine learning.

[0083] When the driving assistance device 010 is turned on, it always collects time series data including the normal range of the following operation acquired during normal driving from the vehicle sensor 002 and the external sensor 003, and stores the plurality of time series data in memory as training data. By setting filtering conditions in advance when collecting the data, it is possible to reduce the amount of memory used.

[0084] The narrowing condition may be, for example, a condition that the collision risk between the vehicle and the vehicle ahead is equal to or less than a certain threshold (a non-dangerous situation). The collision risk is calculated in the same manner as in the explanation of the speed control unit 019 in the first embodiment. For example, if the TTC is greater than a predetermined time, there is a high possibility that the following operation will be within the normal range.

[0085] Furthermore, the conditions may further include that the driver does not perform any of sudden acceleration / deceleration, sudden steering, blinker operation, and shift operation.

[0086] The following methods may also be included in the conditions.

[0087] First, after the narrowing-down condition is satisfied while the driver is driving and a predetermined time has elapsed (after time-series data of several seconds to several minutes has been accumulated), the audio output device 004 or the image display device 005 outputs a sound or a display to the driver to confirm whether the current driving is within a normal range of following behavior. The driver then listens to or visually confirms the sound or display, and if the following behavior is within the normal range and correct, the driver presses a correct operation button on the image display device 005 to notify the driving assistance device 010 that the driving is correct. When the correct operation button is pressed, it is determined that the driver has given permission to use the time-series data as training data, and the driving assistance device 010 stores the time-series data in a memory as training data. In other words, the condition may include the driver being able to confirm that the current driving of the vehicle is a normal following behavior of the driver relative to the vehicle ahead.

[0088] Furthermore, if the above narrowing-down conditions are met and the driver monitor of the vehicle sensor 002 determines that the driver is not in a state of functional decline (in other words, the driver's functional decline is equal to or greater than a certain value), the driving assistance device 010 stores the time-series data in memory as training data.

[0089] The narrowing-down conditions may be acquired from the cloud, and the driving support device 010 may store the time-series data that meets the narrowing-down conditions in memory as training data.

[0090] The training data may be uploaded to the cloud.

[0091] The narrowing conditions may also include the following methods.

[0092] First, when the driver starts driving, time-series data is collected and uploaded to the cloud. Next, after the driver finishes driving, they access the cloud from an external device such as a smartphone or PC, and select teacher data by specifying time-series data from the external device that shows that the driver is not in a state of impaired function. The selected teacher data is acquired from the cloud by the driving assistance device 010 the next time the device is turned on, and the driving assistance device 010 stores the acquired teacher data in memory.

[0093] Alternatively, the driver may specify the time period and location for uploading to the cloud and select the upload conditions on an external terminal before starting to drive. The upload conditions may be conditions such as the collision risk between the vehicle and the vehicle ahead being below a certain threshold (a non-dangerous situation), or the driver not performing sudden acceleration / deceleration, sudden steering, turn of the blinker, or shift operation.

[0094] (Follow-up operation normal learning unit 016) The normal following behavior learning unit 016 uses multiple pieces of training data stored in memory as the correct answer and uses techniques such as deep learning to learn the normal range of following behavior using time series data acquired from the vehicle sensor 002 and the external sensor 003 as input.

[0095] That is, the normal following operation learning unit 016 A feature value indicating that the risk of collision with the vehicle ahead is below a certain value (a non-dangerous situation), A feature amount when the driver of the vehicle does not perform any of sudden acceleration / deceleration, sudden steering, blinker operation, and shift operation, or At least one of the feature quantities obtained when the driver of the vehicle can confirm that the current driving of the vehicle is a normal following operation of the vehicle ahead is used as training data (correct learning data), and the normal range of following operation is learned.

[0096] Furthermore, the normal following operation learning unit 016 calculates the frequency, amplitude, and bandwidth of the relative speed by a method using a fast Fourier transform (FFT) or the like that calculates the fluctuation of the following operation in the normal following operation learning unit 016 described in Example 1 from a plurality of teacher data stored in a memory. These numerical values ​​are interpreted as teacher data, and a method such as deep learning is used to learn the normal ranges of the frequency, amplitude, and bandwidth of the relative speed, in other words, a predetermined range of one or more of the frequency, amplitude, and bandwidth of the relative speed.

[0097] In addition, the normal following operation learning unit 016 may calculate average values ​​of THW, relative speed, inter-vehicle distance, etc. from multiple pieces of teacher data stored in the memory, interpret these values ​​as teacher data, and learn normal ranges of the average values ​​of THW, relative speed, inter-vehicle distance, etc., in other words, predetermined threshold values ​​of the average values ​​of THW, relative speed, inter-vehicle distance, etc.

[0098] That is, the normal following operation learning unit 016 A feature value indicating that the risk of collision with the vehicle ahead is below a certain value (a non-dangerous situation), A feature amount when the driver of the vehicle does not perform any of sudden acceleration / deceleration, sudden steering, blinker operation, and shift operation, or At least one of the feature quantities obtained when the driver of the vehicle is able to confirm that the current driving of the vehicle is a normal following operation of the driver of the vehicle in front is used as training data (correct learning data), and one or more predetermined thresholds of frequency, amplitude, or bandwidth, or a predetermined threshold of the average value of the change in the relative relationship between the vehicle in front and the vehicle in front (relative speed, vehicle distance, THW, etc.) are learned.

[0099] The normal following behavior learning unit 016 narrows down the teacher data under the condition that the driver's functional decline is equal to or greater than a certain value (not in a state of functional decline) as determined by the driver monitor, and learns the normal range of the following behavior. Alternatively, it acquires narrowing down conditions for the teacher data from the cloud, narrows down the teacher data based on the narrowing down conditions, and learns the normal range of the following behavior. Alternatively, it narrows down the teacher data based on any one of the time, place, and conditions specified by the driver, and learns the normal range of the following behavior.

[0100] Furthermore, the normal following operation learning unit 016 learns the normal range of the following operation from the teacher data acquired from the cloud.

[0101] [Effect description] As described above, by efficiently preparing training data and using it to learn the normal range of a driver's following behavior, it is possible to automate or semi-automate responses to the driver's habits and characteristics, thereby reducing false alarms due to a driver's impaired function.

[0102] <Example 3: Determining abnormalities in following operation by detecting sway using lane marking information> The configuration diagram of the driving support device according to the embodiment of the present invention is the same as that of the first embodiment, and only the cases where different processes are performed in each configuration of FIG. 1 will be described below.

[0103] [Configuration explanation] (Follow-up operation normal learning unit 016) The normal following operation learning unit 016 has a normal lane keeping operation learning unit, and learns the normal range of fluctuation in lane keeping operation when the driver of the host vehicle is keeping the vehicle within the lane.

[0104] Fluctuations in lane-keeping behavior refer to the relative movement between the host vehicle and a marking line when the driver of the host vehicle manually keeps the vehicle within the lane. In this embodiment, the relative movement refers to changes in the state of the host vehicle (calculated based on predetermined data) detected over time by the host vehicle sensor 002 or changes in the state between the host vehicle and a marking line detected by the external sensor 003. Specifically, changes in the state of the host vehicle include changes in the steering angle and yaw rate. Furthermore, changes in the state between the host vehicle and a marking line include changes in the relative lateral position of the host vehicle and the marking line.

[0105] Then, the normal lane following operation learning unit 016 determines that the current lane keeping operation is normal if it is within the normal range based on the learned normal range of the lane keeping operation. Note that the normal range refers to the lane keeping operation in a state where the driver of the vehicle is paying attention to driving and keeping the vehicle within the lane.

[0106] In this embodiment, it is assumed that the normal range of lane keeping operation is stored in the non-volatile memory of the driving assistance device 010 as a parameter based on a predetermined threshold value learned in advance theoretically, and the parameter is read and used when the driving assistance device 010 is started. Then, the current lane keeping operation of the vehicle is calculated from changes in the state of the vehicle and changes in the state between the vehicle and the lane markings, and by comparing this with the learned parameters in the normal range, it is determined whether the lane keeping operation is normal.

[0107] To calculate the current lane-keeping operation, for example, a method is used in which the frequency, amplitude, and bandwidth that represent the tendency of change in the relative lateral position with respect to the lane marking are calculated. The calculation method uses techniques such as fast Fourier transform (FFT), similar to the method for lane-following operation in the first embodiment, so details are omitted. Furthermore, similar to the method for lane-following operation in the first embodiment, the average value of change in the relative lateral position with respect to the lane marking may also be used for determination.

[0108] That is, the normal lane following operation learning unit 016 (its normal lane keeping operation learning unit) calculates one or more of frequency, amplitude, or bandwidth by frequency converting the change in the relative relationship between the vehicle and the lane marking (such as a change in the relative lateral position with respect to the lane marking), and learns it as normal (within the normal range) if one or more of the calculated frequency, amplitude, or bandwidth is within a predetermined threshold range, or if the average value of the change in the relative relationship between the vehicle and the lane marking (such as a change in the relative lateral position with respect to the lane marking) is within a predetermined threshold range.

[0109] When the driving assistance device 010 is turned on, it constantly collects time series data including the normal range of lane keeping operation acquired during normal driving by the driver from the vehicle sensor 002 and the external sensor 003, and stores the plurality of time series data in memory as training data. By setting filtering conditions in advance when collecting the data, it is possible to reduce memory usage.

[0110] The narrowing condition may be, for example, a condition in which the possibility of deviation between the vehicle and the lane marking is below a certain threshold (a non-dangerous situation). The deviation possibility may be, for example, if the time it takes for the vehicle to deviate from the lane marking, based on the time to line crossing (TTLC), is greater than a predetermined time, then the following operation is likely to be within the normal range. The TTLC can be calculated by dividing the lateral speed of the vehicle by the relative lateral position of the lane marking.

[0111] Other narrowing down conditions are the same as those in the tracking operation of the second embodiment, so details will be omitted.

[0112] The lane-keeping normal learning unit, like the following operation normal learning unit 016, uses multiple pieces of teacher data stored in the memory as correct answers and inputs time-series data acquired from the vehicle sensor 002 and the external sensor 003, and uses techniques such as deep learning to learn the normal range of following operation.

[0113] That is, the normal following operation learning unit 016 (its normal lane keeping operation learning unit) uses as teacher data (learning data that is the correct answer) feature amounts that indicate that the possibility of deviation from a change in the relative relationship between the vehicle and the lane marking (such as a change in the relative lateral position with respect to the lane marking) is below a certain value (a situation that is not dangerous), and learns the normal range of following operation.

[0114] [Effect description] As a result, the driver's impaired function can be detected regardless of whether there is a vehicle ahead, which makes it possible to expand the scope of application.

[0115] <Summary of Examples 1 to 3> As described above, the driving assistance device 010 of this embodiment includes a vehicle-in-front recognition unit 014 that detects a vehicle in front that is traveling ahead of the vehicle in question; a normality of following behavior learning unit 016 that learns whether the tendency for fluctuations in the following behavior of the driver of the vehicle in question regarding the vehicle in front is within a normal range, based on at least one feature of a change in the state of the vehicle in question and a change in the state between the vehicle in question and the vehicle in front, which are calculated based on predetermined data detected in a time series; a follow-up behavior abnormality determination unit 017 that determines that the current following behavior is abnormal if the fluctuations in the current following behavior differ by a certain amount or more from the normal range; a follow-up behavior abnormality determination suppression unit 015 that predicts a transient state in which it is not possible to determine whether the fluctuations in the current following behavior are normal or abnormal, and suppresses the determination of the follow-up behavior abnormality determination unit 017; and an alarm control unit 018 that issues an alarm when the current following behavior is determined to be abnormal.

[0116] That is, the driving support device 010 of this embodiment detects a tendency for fluctuations (unusual fluctuations) in following the vehicle ahead, detects a functional decline state of the driver, and issues a warning.

[0117] According to this embodiment, by detecting the signs of the driver entering a state of impaired function, it is possible to warn the driver at an earlier stage and prevent or mitigate a collision with an obstacle.

[0118] It should be noted that the present invention is not limited to the above-described embodiment, and includes various modifications. For example, the above-described embodiment has been described in detail to clearly explain the present invention, and the present invention is not necessarily limited to an embodiment having all of the described configurations.

[0119] Furthermore, the above-described configurations, functions, processing units, processing means, etc. may be partially or entirely implemented in hardware, for example, by designing them as integrated circuits. The above-described configurations, functions, etc. may also be implemented in software, with a processor interpreting and executing a program that implements each function. Information such as the programs, tables, and files that implement each function can be stored in a memory, a storage device such as a hard disk or SSD (Solid State Drive), or a recording medium such as an IC card, SD card, or DVD.

[0120] In addition, the control lines and information lines shown are those that are considered necessary for the explanation, and do not necessarily show all the control lines and information lines in the product. In reality, it can be assumed that almost all components are interconnected. [Explanation of symbols]

[0121] 001 Vehicle (own vehicle) 002 Vehicle sensor 003 External Sensor 004 Audio output device 005 Image display device 006 Accelerator 007 Reduction device 008 Car Navigation System (CNS) 010 Driving support device 013 Surrounding Environment Awareness Department 014 Front vehicle recognition unit 015 Tracking operation abnormality determination suppression unit 016 Normal learning of following operation 017 Tracking operation abnormality determination unit 018 Alarm control section 019 Speed ​​control section

Claims

1. A learning device that learns the normal range of the driver's following behavior of the vehicle ahead, An acquisition unit that acquires feature quantities based on time-series data of the vehicle's own sensors and external sensors acquired in time series, A memory unit for storing training data used in supervised learning, A training data generation unit, when predetermined filtering conditions are met, collects time-series data from the time-series data acquired during normal operation that includes the portion within the normal range of the tracking operation, and stores it in the storage unit as training data. A learning device comprising: a learning unit that learns the normal range of the tracking operation by taking as input time-series data acquired by the acquisition unit, with multiple training data stored in the storage unit as correct answers.

2. A learning device according to claim 1, The learning device is characterized in that the aforementioned narrowing condition includes the fact that the collision risk between the vehicle itself and the vehicle ahead is below a predetermined threshold.

3. A learning device according to claim 2, The learning device is characterized in that the collision risk is determined by collecting training data, assuming that the tracking operation is likely to be within the normal range when the TTC is greater than a predetermined time.

4. A learning device according to claim 1, The aforementioned filtering conditions are characterized in that the driver has not performed any of the following actions: sudden acceleration / deceleration, sudden steering, turn signal operation, or shift operation.

5. A learning device according to claim 1, The aforementioned vehicle sensors include a driver monitor, The aforementioned filtering condition is a learning device characterized in that the driver is not in a state of impaired function, as determined by the driver monitor.

6. A learning device according to claim 1, The training data generation unit, after the filtering conditions are met and a predetermined time has elapsed, outputs a confirmation using an audio output device or image display device installed in the vehicle to verify whether the current operation is a follow operation within the normal range. A learning device characterized by storing the time-series data as training data in the storage unit based on the response from the driver.

7. A learning device according to claim 1, The learning device is characterized in that the aforementioned training data is uploaded to the cloud.

8. A learning device according to claim 7, A learning device characterized in that, after the driver has finished driving, the driver accesses the cloud from an external terminal and selects the training data by specifying time-series data in which the driver is not in a state of functional degradation.

9. A learning device according to claim 1, The training data generation unit narrows down the training data based on one of the time, place, or conditions specified by the driver. The learning device is characterized in that the learning unit learns the normal range based on the narrowed-down training data.

10. A learning device according to claim 1, The learning device is characterized in that the learning unit reads at least one of the frequency, amplitude, and bandwidth obtained by frequency conversion to calculate the fluctuation of the tracking operation from a plurality of training data as training data, and learns the normal range of the frequency, amplitude, and bandwidth by deep learning.

11. A learning device according to claim 1, The learning device is characterized by the learning unit calculating at least one average value of THW, relative speed, and inter-vehicle distance from a plurality of training data, and learning the normal range of the average value.

12. A learning method for learning the normal range of the driver's following behavior of the vehicle in front, Computers We acquire feature quantities based on time-series data from our own vehicle sensors and external sensors, which are acquired in a time-series manner. The training data used for supervised learning is stored in the memory unit. When the predetermined filtering conditions are met, time-series data obtained during normal operation, including the portion that falls within the normal range of the tracking operation, is collected and stored in the memory unit as training data. A learning method characterized by learning the normal range of the tracking operation by using a plurality of training data stored in the memory unit as correct answers and the acquired time-series data as input.

Images