JP2751341B2 - Road condition recognition device for vehicles - Google Patents

Description

The present invention relates to a vehicle road condition recognition device, and more particularly, to a vehicle road condition recognition device.
The present invention relates to a vehicle road condition recognition device that recognizes the condition of a road on which a vehicle is traveling as one of a plurality of road conditions (e.g., urban roads, highways, and congested roads) that are classified in advance.

2. Description of the Related Art Conventionally, there has been proposed an apparatus that recognizes a state of a road on which a vehicle is traveling and provides the information as traveling information to a driver or provides the information to various controls of the vehicle. For example, an attempt has been made to recognize a road condition such as a traffic jam or an urban area by inputting an image of the outside world using an imaging device and performing image processing such as binarization based on the image. Further, there has been proposed a sensor that detects a distance between vehicles before and after a vehicle by a distance sensor using a laser or the like and uses the distance to recognize road conditions.

[Problems to be Solved by the Invention] However, in order to recognize road conditions by image processing in a running vehicle, basically, moving images must be processed. However, it is extremely difficult to realize the above-mentioned apparatus due to restrictions on arithmetic processing time and cost. In addition, in the case of trying to recognize the situation such as traffic congestion from the vehicle distance or the like, a new sensor is required, and it is difficult to accurately recognize the inter-vehicle distance and the like because there is a large individual difference between drivers. There was a problem.

That is, at present, it is extremely difficult to accurately recognize the state of the road on which the vehicle travels while traveling.

Therefore, the present invention describes the condition of the road on which the vehicle is traveling,
An object of the present invention is to provide a vehicle road condition recognizing device capable of performing high-speed and accurate recognition.

[Means for Solving the Problems] That is, according to the present invention for achieving the above-described object, as shown in FIG. 1, the state of a road on which a vehicle is traveling is defined as one of a plurality of road conditions classified in advance. A vehicle road condition recognition device for recognizing, comprising: a driving data output unit M1 for detecting a plurality of parameters reflecting a driving state of a vehicle and outputting data for each parameter; and An input layer IL, which is a set of input units Im (m = 1, 2, ‥‥) to be input, and a plurality of intermediate units Mn (n = 1, 2) at least partially connected to each input unit Im of the input layer IL. 2, ‥‥) and an output unit Op (p = 1, 2, ‥) coupled to at least a part of the plurality of intermediate units Mn and provided for the plurality of road conditions. Neural network M2 consisting of output layer OL, which is a set of i) A road condition input means M3 for designating a currently traveling road condition from outside; and, when a currently traveling road condition is designated via the road condition input means M3, an output layer of the neural network M2. Learning means M4 for calculating the strength of coupling between the units between the layers so that the output from each unit Op of the OL becomes the desired output representing the specified road condition. The main purpose is to recognize road conditions for vehicles.

[Operation] In the vehicle road condition recognition device of the present invention configured as described above, first, the traveling data output means M1 outputs the traveling state of the vehicle (for example, the accelerator pedal depression amount, vehicle speed, brake operation state, steering wheel steering angle). ) Is detected and data is output for each parameter.

Next, each data of each parameter output from the traveling data output means M1 is input to each input unit Im of the input layer IL of the neural network M2. Each input unit Im of the input layer IL is composed of a plurality of intermediate units Mn of the intermediate layer ML.
And at least a part of the plurality of intermediate units Mn is connected to a plurality of output units Op of the output layer OL. Each output unit Op of the output layer OL is provided corresponding to a plurality of road conditions (for example, an urban road, an expressway, a congested road, a mountain road, a snowy road, an icy road, and the like). When input to the input layer IL of the net N, a signal representing the road condition is output from each output unit Op of the output layer OL.

On the other hand, when the currently traveling road condition is designated via the road condition input means M3, the learning means M4 operates, and the designated road condition is used as so-called teacher data, and each of the output layers OL of the neural network M2 is output. A neural network is used so that the output from the unit Op is an output representing the specified road condition.
A so-called learning operation of calculating the strength of the connection between the units between the layers of M2 is executed. In other words, road condition input means
By specifying the current road condition via M3, the strength of the connection between each unit between the layers of the neural network M2 has corresponded to the road condition during vehicle driving from each unit Op of the output layer OL. Updated to get output.

For this reason, as the learning means M4 learns the neural network M2, a desired output corresponding to the actual road condition can be obtained from each output unit Op of the output layer OL of the neural network M2.

That is, in the present invention, by the neural network M2,
Based on a plurality of parameters reflecting the traveling state of the vehicle, the state of the road on which the vehicle is currently traveling is recognized as one of a plurality of road conditions classified in advance, and the recognition accuracy of the learning means M4 is determined. Improve by learning operation.

Here, the data on each parameter output from the traveling data output means M1 may be a predetermined number of time-series data, or may be data of an average value, a change amount or a change frequency in a predetermined time.

Further, the neural network M2 can adopt various configurations such as a hierarchical type, a free associative type, and a mutual connection type. In the case of the hierarchical type, the intermediate layer is not limited to one layer, and two or more layers may be used. It may be configured to have a neural network
You can add feedback coupling to M2.

Further, as a learning rule for calculating the strength of the connection between the units between the layers by the learning means M4, a well-known algorithm such as a back propagation learning rule (back propagation) can be applied. , A method such as a recirculation algorithm or recursive back propagation can be used.

Embodiment An embodiment of the present invention will be described below with reference to the drawings.

First, FIG. 2 is a schematic configuration diagram of a vehicle road condition recognition apparatus as an embodiment.

As shown in the figure, the vehicle road condition recognition device of the present embodiment includes a sensor group that is disposed on the internal combustion engine 1 or the transmission 3 as a vehicle drive source and detects parameters related to the traveling state of the vehicle. A display device 5 for displaying a recognition result of
It comprises a road condition recognition circuit 7 for recognizing a road condition, and a select switch 9 for designating a road condition in which the vehicle is currently traveling.

First, the sensor group will be described. As a sensor for detecting the running state of the vehicle, the operation amount AC of the accelerator pedal 10 is used.
An accelerator sensor 11 for detecting C, a steering angle sensor 14 for detecting an operation amount (ie, a steering angle) H of a steering wheel 12, a brake switch 18 for detecting on / off of a brake pedal 16, and a vehicle speed sensor 19 for detecting a vehicle speed V Etc.

The accelerator sensor 11 that outputs an analog signal corresponding to the accelerator operation amount ACC is also connected to the throttle control circuit 21. The throttle control circuit 21 sets a control characteristic of the throttle opening θ with respect to the accelerator operation amount ACC according to the road condition in which the vehicle is currently traveling, and controls the throttle opening θ based on the accelerator operation amount ACC according to the control characteristics. In addition to the detection signal from the accelerator sensor 11,
A detection signal from the throttle sensor 23 for detecting the opening degree (throttle opening degree) θ of the throttle valve 22 and an output signal from the road condition recognition circuit 7 are input, and the throttle opening degree is controlled via the motor 24. It is configured.

The steering angle sensor 14 is provided on the rotating shaft 12a of the steering wheel 12, and outputs a pulse signal corresponding to the steering angle H of the steering wheel 12 using a photo interrupter. The brake switch 18 is a sensor that detects whether the brake pedal 16 is on or off. The vehicle speed sensor 19 is provided on the output shaft 3a of the transmission 3 and outputs pulse signals at intervals corresponding to the vehicle speed.

A road condition recognition circuit 7 for inputting signals from these sensor groups and recognizing road conditions includes a well-known CPU 31, ROM 32,
A timer 35 for generating an interrupt signal and the like, an input port 37 for inputting signals from the sensor group and the select switch 9, a recognition result to the display device 5, a throttle control circuit, and the like. An output port 38 for outputting to 21 and the like is provided, and these are interconnected via a bus 39. The RAM 33 is configured as a so-called backup RAM backed up by a battery so that the latest data used for road recognition is not lost when the power switch of the vehicle is turned off.

Next, the input port 37 and the sensor group
In the case of 18, it is connected directly, but in the case of the steering angle sensor 14 and the like, it is connected via signal conversion circuits 41, 43, 45. A signal conversion circuit 41 that processes a signal from the steering angle sensor 14 includes:
It incorporates an up / down counter, counts a pulse train output from the steering angle sensor 14, and converts the pulse train into a signal corresponding to the steering angle. Further, the signal conversion circuit 43 that processes the signal from the accelerator sensor 11 is configured as an A / D converter that converts an analog signal into a digital signal having a predetermined number of bits. Further, the signal conversion circuit 45 for processing the signal from the vehicle speed sensor 19 has a built-in self-propelled timer, and detects the time interval of the pulse train output from the vehicle speed sensor 19 with the timer, which corresponds to the vehicle speed V. Is converted into data to be processed. The data converted by these signal conversion circuits 41, 43, and 45 is input to the road condition recognition circuit 7 through the input port 37.
And is used in the process of recognizing road conditions.

Select switch 9 connected to input port 37
Consists of a plurality of switches corresponding to each road condition recognized by the road condition recognizing circuit 7 (in the present embodiment, an urban road, an expressway, a congested road, an uphill road, a snowy road). By operating, the road condition recognition circuit 7 can be instructed on the current road condition.

The road situation recognition circuit 7 is realized by a so-called neural network, which is realized by software in existing hardware of a sequential processing type.

Hereinafter, the processing of data from each sensor group and the select switch 9 and the configuration of the neural network will be described with reference to FIG.

Signals such as vehicle speed data and accelerator operation amount data are read at predetermined intervals in a data capturing unit SDi (i = 1, 2, 3, 4), and a time-series data expanding unit TDSi (i = 1, 2, 3, Four)
Is developed into time-series data at predetermined time intervals.
The time-series data development unit TDSi has a configuration in which data is shifted every unit time to generate a time-series data by adding a unit time delay to the data.

The neural network NN is of a hierarchical type, and has an input layer composed of input units Im provided corresponding to each data expanded in time series, and an intermediate unit Mn provided almost as many as the input units. And an output layer composed of output units Op provided by the number corresponding to the road conditions to be recognized. Although the strength of the connection between the units is learned in advance by the back propagation method, in the present embodiment, the learning is further advanced by designating the current road condition via the select switch 9. Thus, the recognition accuracy of road conditions can be further improved.

Next, the input data from the select switch 9 is input to the learning signal generator GH. The learning signal generation unit GH generates a learning ON-OFF signal that is turned on when any one of the select switches 9 is turned on (that is, when a road condition is designated), and selects the learning signal. When the road condition is designated by the switch 9, a signal to be output from each output unit Op of the neural network NN under the road condition is generated as teacher data Tp. The learning signal generating unit GH is realized as an operation of the input port 37. By operating the select switch 9, the learning ON-OFF signal and the teacher data T
p is generated.

The teacher data and the learning ON-OFF signal generated by the learning signal generator GH are input to the learning controller GS. When the learning ON-OFF signal is in the OFF state, the learning control unit GS outputs the output value from each output unit Op of the neural network NN as it is as data representing the recognition result of the road condition, and outputs the learning value from the learning signal generation unit GH. When the learning ON-OFF signal is turned on, a deviation between the output value from each output unit Op of the neural network and the corresponding teacher data is generated as error data.
A learning operation for updating the strength of the connection between the units of the NN is executed. In the learning control unit GS, at the time of executing the learning operation, the teacher data generated by the learning signal generating unit GH is output as data representing the recognition result of the road condition.

In the learning control unit GS, learning of the neural network NN is performed by a back propagation method. As is well known, the learning rule based on back propagation learns the strength (coupling amount) ω of coupling between units so that the error between the actual output and the ideal output of each output unit Op is minimized. Since the least squares error method is used, the output value of the output unit Op corresponding to the road condition specified by the select switch 9 is set to the value 1 in accordance with the learning rule in this embodiment. Is generated with the output value of the output unit Op having a value of 0, and the strength of the connection (the amount of connection) between the units so that the error between the output value of each output unit Op and this teacher data is minimized. Learn ω.

On the other hand, the time-series data expanded by the time-series data expansion unit TDSi is input to each input unit lm of the input layer of the neural network NN. FIG. 4 schematically shows this state. is there. The time-series data is updated by deleting one oldest data every unit time and adding one newest data.

Summarizing the above processing simply, this road condition recognition device for a vehicle inputs time-series data of each input signal during traveling for a certain period of time, calculates the output, recognizes the road condition, and selects the road condition. When a road condition is designated via the switch 9, the strength of the connection between the layers in the neural network NN set in advance is learned and updated by the back propagation method. In the present embodiment, the previously learned coupling strength is stored in the ROM 32 as the coupling amount ω, and the coupling amount ω updated by learning during vehicle running is stored in the RAM 33 as the latest data used for road recognition. I do.

Hereinafter, the road condition recognition processing actually executed by the road condition recognition circuit 7 for recognizing the road condition will be described with reference to the flowchart of FIG.

This process is started every 10 [msec] as a 10 msec interrupt routine, and first outputs a command to capture the data Di in step 100. Here, the data Di is the vehicle speed data V output from the vehicle speed sensor 19, the accelerator operation amount data ACC output from the accelerator sensor 11, the on / off signal B output from the brake switch 18, and the steering angle data output from the steering angle sensor 14. H. When such an instruction for capturing the data Di is output, the process proceeds to step 105, where the signal conversion circuits 41, 43, and 45 that have received the instructions perform the above-described processing such as A / D conversion and complete the data capture. Wait until.

When the data fetching is completed, the process proceeds to the subsequent step 110, in which a process of shifting all the input time-series data TDm input so far and stored in a predetermined area of the RAM 33 is performed.
As a result, all data TDm is delayed by the unit time To (interruption interval 10 [msec] in this embodiment). Next, in step 115, a process of loading each latest data fetched in steps 100 and 110 into a memory for storing data of the latest time in an area where all data TDm is stored is performed.

Through the above processing, the update of all the time series data TDm is completed, and thereafter, the processing as the neural network NN is executed.

That is, first, in step 120, the output value IDm of each unit Im of the input layer is calculated. In this calculation, the time series data TDm of each input data Di is multiplied by a predetermined coupling amount ωm to obtain an accumulated value NETm by the following equation (1). Output value ID as a jigmoid function
This is done by finding m.

NETm = ωm · TDm ... (1) Note that the jigmoid function of the above equation (2) is the cumulative number NETm = 0.
When the function value becomes 0 and the cumulative value NETm reaches the maximum value, the function value becomes 1 and when the cumulative value NETm becomes an intermediate value in the domain, the phase value Sm becomes 0.5.
Is determined. In the above equations (1) and (2) and the equations described below, ω is a coupling amount indicating the strength of coupling between the layers, and ωm is a value between the input data and the corresponding input unit Im. , Ωnm is between the input unit Im and the intermediate unit Mn, ωpn is the intermediate unit Mn and the output unit
It is the amount of binding between Owp. As described above, these coupling amounts are values updated by a learning operation to be described later by a back propagation method.

Next, in step 125, a process of calculating the output value MDn of each unit Mn of the intermediate layer according to the following equations (3) and (4) is performed. The calculation of this output value MDn is based on the input unit lm
Is performed in the same manner as the calculation of the output value lDm.

Next, in step 130, the output value ODp of each unit Op of the output layer is calculated according to the following equations (5) and (6). This output value ODp is calculated by the intermediate unit M
The calculation is performed in the same manner as the calculation of the output value MDn of n.

When the output value ODp of each unit Op of the output layer is calculated in this manner, the process proceeds to step 135, where the select switch 9
It is determined whether the learning ON-OFF signal is in the ON state by the operation of. If the learning ON-OFF signal is in the OFF state, that is, if the road condition has not been designated via the select switch 9, the process proceeds to step 140, and the process proceeds to step 130.
The output value ODp of each unit Op of the output layer calculated in the above is output via the output port 38.

Note that this output value ODp is a recognition result by the neural network NN. If a pattern at the time of learning or a pattern similar thereto is input, the corresponding output value ODp is approximately set to the value 1 and the other output values ODp Is approximately 0. Therefore, this output value ODp becomes the road condition recognition result as it is.

Next, when it is determined in step 135 that the learning ON-OFF signal is in the ON state, the process proceeds to step 145, where the teacher data Tp generated according to the operation state of the select switch 9 at the input port 37 is output. Read. In the subsequent step 145, the read teacher data Tp is output to the output port 3
Output via 8. That is, when the road condition is designated by operating the select switch 9, the teacher data Tp representing the designated road condition is output as it is.

When the output of the teacher data Tp is completed in this way, the learning processing of the neural network NN is performed by the processing of the subsequent steps 155 and thereafter.

That is, first, in step 155, the error value δp of each unit Op of the output layer is calculated as the teacher data Tp, the output value ODp of each unit Op of the output layer obtained in step 130, and the cumulative value of each unit Op of the output layer obtained in step 130. Differential value fp 'of a function representing the input / output relationship of each unit Op in the output layer using the value NETp as a parameter
Based on (NETp), it is calculated according to the following equation (7).

δp = (Tp−ODp) · fp ′ (NETp) (7) Next, at step 160, the error value δn of each unit Mn of the intermediate layer is calculated by each unit Mn of the intermediate layer obtained at step 125.
The differential value fn '(NETn) of the function representing the input / output relationship of each unit Mn in the intermediate layer, the error value δp of each unit Op in the output layer obtained in step 155, and the unit Mn in the intermediate layer Ω between the output layer and each unit Op
Based on pn, it is calculated according to the following equation (8).

Next, in step 165, the error value δm of each unit Im of the input layer is calculated using the accumulated value NETm of each unit Im of the input layer obtained in step 120 as a parameter.
The derivative fm '(NETm) of the function representing the input / output relationship of Mn, the error value δn of each unit Mn of the intermediate layer obtained in step 160,
And the amount of coupling ω nm between each unit lm of the input layer and each unit Mn of the intermediate layer is calculated according to the following equation (9).

In the following step 170, each unit (lm,
Mn, Op) and the correction amount △ ω (△ ωm, △ ωnm, △ ωpn) of the coupling amount ω (ωm, ωnm, ωpn) with the learning constant η,
Each error value δ (δm, δn, δp) of each unit (Im, Mn, Op) of each layer obtained in steps 155 to 165 and each unit (Im, Mn) of each layer obtained in steps 120 to 130 , Op) based on the output value D (IDm, MDn, ODp) using the following equation (10) Δω = η · δ · D (10). Using the quantity △ ω (△ ωm, △ ωnm, △ ωpn), the coupling amount ω (ωm, ω) with the input data of each unit (Im, Mn, Op) of each layer
By correcting (= ω + △ ω) the coupling amount ω (ωm, ωnm, ωpn).

The learning rules based on back propagation are described in detail in Kazuyuki Aihara, "Neural Computer" (published by Tokyo Denki University Press), Hideki Aso, "Neural Network Information Processing" (published by Sangyo Tosho Co., Ltd.) Here, detailed description is omitted.

As described above, in the road situation recognition circuit 7 of the present embodiment, when the currently traveling road situation is designated via the select switch 9, the output value of each unit Op of the output layer of the neural network NN is designated. The amount of connection of each layer of the neural network NN is learned so as to be an output value corresponding to the road condition that has been obtained, and if the road condition is not specified, the output value of each unit Op of the output layer obtained based on the learning result is calculated as Is output as a recognition result.

Next, the display device 5 that receives the output signal from the road condition recognition circuit 7 is configured to display the recognized road surface condition by lighting a lamp corresponding to the output value ODp that is equal to or greater than a predetermined threshold value. I have. For this reason, if the recognition is not performed properly in a transitional road condition, none of the output values ODp exceeds the recognition threshold value, so that none of the lamps on the display device 5 is turned on.

The output value ODp is also output to the throttle control circuit 21, and the throttle control circuit 21
The relationship between the accelerator operation amount and the throttle opening θ is switched based on the recognition of the road condition reflected in ODp.

That is, as shown in FIG. 6, the throttle control circuit 21 has a throttle opening characteristic for an urban road, which is used for calculating the target throttle opening θO from the accelerator operation amount ACC, with respect to the throttle opening θ. The normal characteristics (shown by the dotted line in the figure) to obtain almost constant accelerator sensitivity
Similarly, as a throttle opening characteristic on a highway, the accelerator sensitivity is reduced on the side where the throttle opening θ is small in order to improve operability in the normal range, and driving torque is required in the middle part of the throttle opening θ. The characteristic of increasing the accelerator sensitivity (indicated by the solid line in the figure) so that the engine torque can be increased quickly in such a case is the same as the throttle opening characteristic for uphill roads with a small accelerator operation amount.
The characteristic of increasing the accelerator sensitivity so that the desired drive torque can be obtained quickly with the ACC opening the throttle valve wide (shown by the dash-dot line in the figure) is the same as the throttle opening characteristic on congested roads. The characteristic (shown by a two-dot chain line in the figure) in which the accelerator sensitivity on the side with a small throttle opening is extremely low is stored in advance. Based on an output signal from the road condition recognition device 14, a throttle opening characteristic corresponding to the road condition is selected from the plurality of throttle opening characteristics, and a target throttle opening θO is determined based on the throttle opening characteristic. The throttle opening control is executed in such a procedure as to control the throttle opening θ.

The output ODp corresponding to the snowy road is the throttle control circuit 2.
In No. 1, it is not used and is output to an anti-skid control device or a traction control device (not shown).

As described above, according to the vehicle road condition recognition device of the present embodiment, by using the neural network NN, the vehicle travels from the vehicle speed V, the accelerator operation amount ACC, the brake data B, and the steering angle data H. It is possible to easily recognize the state of the road on which the vehicle is traveling. Therefore, it is possible to recognize a road condition at a very high speed as compared with a conventional method for recognizing a road condition by performing image recognition.

Further, in the present embodiment, by specifying the current road condition via the select switch 9, the coupling amount ω of each layer of the neural network NN can be learned. Therefore, by this learning operation, the recognition error of the road condition can be corrected and the recognition accuracy of the road condition can be improved.

Further, it is not necessary to use a new image pickup device as in image recognition, and a conventionally used vehicle control signal can be diverted as it is, and advantages such as cost reduction and simplification of the system can be obtained.

Further, since the neural network NN is used, there is no need to make a rule for the determination criteria for recognizing road conditions. In addition, in the present embodiment, learning is easy because a hierarchical neural network is used.

Further, with the adoption of the hierarchical neural network NN, there is an advantage that an interpolative recognition result can be obtained in a transient road condition. That is, when the road conditions change, for example, in a transitional state where traffic congestion starts from normal driving on a city road, an intermediate output between the two road conditions is obtained, and the determination of the road conditions is greatly erroneous. There is no.

In this embodiment, 10 seconds of time-series data (1000 points for each data) updated every 10 [msec] is used for data such as the accelerator operation and the brake on / off operation of the driver. Therefore, there is an advantage that recognition accuracy is extremely high.

Here, in the above embodiment, the time series data TDm is used as the input data to the neural network NN. However, for example, the vehicle speed data V, the accelerator operation amount data ACC, and the like for a predetermined time (100 [msec. ]) May be obtained and input to the neural network NN as input data. In this case, in addition to the effects of the above-described embodiment, the number of input data to the neural network NN can be reduced by averaging the data, and the effects such as simplification of the device configuration and improvement of the processing speed can be obtained. be able to. Further, by taking the average value, it becomes strong against noise superimposed on the input data, and the possibility of erroneous recognition can be reduced.

Further, as input data to the neural network NN, a vehicle speed data V, an accelerator operation amount data ACC, a change amount of the steering angle H within a predetermined time, and the number of times of turning on / off the brake switch 18 are obtained. You may make it input into NN. In this case, in addition to the effects of the above embodiment, it is possible to reduce the amount of input data to the neural network NN by handling the amount of change in data, thereby simplifying the device configuration and improving the processing speed. The effect can be obtained. Further, by extracting the change, it becomes possible to recognize the change in the road condition with good responsiveness.

Further, in the above embodiment, the road conditions recognized by the neural network NN include an urban road, an expressway, an uphill road,
Although five types of traffic congestion roads and snowy roads are set, road conditions to be recognized may be further subdivided into, for example, snowy roads and icy roads or mountain roads going up and down.

[Effects of the Invention] As described in detail above, according to the vehicle road condition recognition device of the present invention, the road condition is represented by parameters representing the running state of the vehicle, such as vehicle speed, accelerator operation amount, brake on / off, and steering angle. Therefore, there is an excellent effect that the road condition can be recognized at a very high speed. In addition, since the neural network is used, there is basically no need to formulate a criterion for determining road recognition, there is no need to examine the validity of the criterion, and the development of a recognition device can be shortened. . Furthermore, since the strength of the connection between the units between the layers of the neural network can be learned by designating the current road condition from the outside, the recognition error of the road condition is corrected and the road condition is corrected. The recognition accuracy can be improved. Further, it is not necessary to use an image pickup device as in an apparatus based on image recognition, and the control signal of the vehicle can be diverted as it is, so that the system can be simplified. Therefore,
It also contributes to cost reduction of the equipment.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a basic configuration of the present invention, FIG. 2 is a schematic configuration diagram of a vehicle road condition recognition device as one embodiment of the present invention, and FIG. FIG. 4 is an explanatory diagram showing the relationship between input time-series data and the neural network NN. FIG. 5 is a flowchart showing a road situation recognition process executed by a road situation recognition circuit. Is a graph showing the control characteristics of the throttle opening controlled using the recognition result of the road condition,
It is. M1… running data output means M2, NN… neural network M3… road condition input means, M4… learning means Im… input unit, Mn… intermediate unit Op… output unit TDSi… time series data expansion Unit GS: Learning control unit, GH: Learning signal generation unit 7: Road condition recognition circuit, 9: Select switch 11: Accelerator sensor, 14, Steering angle sensor 18: Brake switch, 19: Vehicle speed Sensor

Continuation of the front page (56) References JP-A-60-255576 (JP, A) JP-A-61-220971 (JP, A) JP-A-64-82133 (JP, A) JP-A-1-232464 (JP) JP-A-1-271888 (JP, A) JP-A-1-320565 (JP, A) JP-A-63-214214 (JP, U) JP-A-63-168813 (JP, U) JP-A 60-193829 (JP, U)

Claims (1)

(57) [Claims] 1. A vehicular road condition recognition device for recognizing a road condition on which a vehicle is traveling as one of a plurality of pre-classified road conditions. Detect
Driving data output means for outputting data for each parameter; an input layer which is a set of input units for inputting each of the data of each parameter; a plurality of input units at least partially connected to each input unit of the input layer A neural network consisting of an intermediate layer, which is a set of intermediate units, and an output layer, which is a set of output units coupled to at least a part of the plurality of intermediate units and provided for the plurality of road conditions; A road condition input means for designating a road condition during the current traveling from the vehicle. When a road condition currently traveling is designated via the road condition input means, an output from each unit of the output layer of the neural network is output. Learning means for calculating the strength of the connection between the units between the layers so that the desired output representing the specified road condition is obtained. A road condition recognition device for a vehicle, comprising: