JP2011014038A - Online risk recognition system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an on-line risk recognition system for improving robustness with respect to various external field environments, when recognizing a risk by autonomously learning experience under actual environment.SOLUTION: A model group obtained by dividing an SOM on one dimensional upper loop into a plurality of models, according to a traveling state is set in parallel according to each category of day and night, or the like. Then, the model to be used is selected, on the basis of information such as a vehicle speed or a steering angle from among the model group selected by a recognition model setting part 5a, so that it is possible to avoid arrangement of the unit in a section where input data are thin. Then, only an image-featured value extracted by a featured value extraction part 4 is input to an SOM model so as to be recognized and learnt by a recognition and learning part 5b, so that it is possible to improve robustness, with respect to data input whose characteristics are different. Furthermore, correlation between a state and a teacher is searched by a risk recognition part 7, and the risk of the state is learnt and recognized. Thus, robustness with respect to various external field environments is improved, at risk recognition.

Description

  The present invention relates to an online risk recognition system that adaptively learns and recognizes risks contained in the external environment of a moving object such as an automobile.

  In recent years, as a preventive safety technology for moving objects such as automobiles, a camera is installed to image the external environment, the captured image is processed to recognize risk information contained in the external environment, and alert the driver Technology has been developed to assist driving or assist driving.

  Such a technique for recognizing danger information is disclosed in, for example, Patent Document 1. The technique of Patent Literature 1 sets a risk parameter for each type and attribute of an object in the environment around the vehicle, and calculates the risk based on the risk parameter.

  In the prior art disclosed in Patent Document 1, factors that lead to danger such as pedestrians, oncoming vehicles, obstacles, white lines, etc. are set, and the risk is recognized based on these factors. Is realized in the form of incorporating risk factors and recognition assumed by the developer into the system in advance.

  However, the actual environment such as the driving environment of automobiles is diverse, such as weather changes, the presence of pedestrians, cars, structures on the road, etc. Furthermore, the number of people driving is also diverse. There is a limit in one of the preset recognition models, and if the factors leading to danger are not recognized with high accuracy, not only the overall risk can be recognized, but also dangerous scenes other than those assumed in advance. There is a problem that cannot be recognized.

  For this reason, in this patent application, the applicant has learned online experience that allows the system to autonomously learn the experience in an actual environment and to recognize the degree of danger corresponding to various external environments. A system is proposed.

JP 2003-81039 A JP 2008-238831 A

  In the technique of Patent Document 2, since the system autonomously learns the experience in the actual environment, it is possible to recognize in accordance with the driving tendency of the user. However, the technique disclosed in Patent Document 2 has a possibility that the learning and recognition efficiency may be reduced due to a difference in location or time zone in an actual environment, and there is room for improvement in improving robustness. It was.

  The present invention has been made in view of the above circumstances, and is an on-line risk recognition that can improve the robustness to various external environments when autonomously learning the experience in an actual environment and recognizing the danger level. The purpose is to provide a system.

  In order to achieve the above object, an online risk recognition system according to the present invention is an online risk learning system that detects an external environment of a mobile object and recognizes a risk included in the external environment in a learning manner. A feature extraction unit that processes detection information to extract multi-dimensional feature values included in the external environment, a recognition model setting unit that sets and sets recognition models for clustering the feature values, and the recognition A state recognition unit that recognizes the multi-dimensional feature quantity as a one-dimensional state using the recognition model set by the model setting unit, and a risk information related to the state recognized by the state recognition unit and the risk level are extracted. A risk recognizing unit that adaptively learns the risk level of the above state based on the correlation with the teacher information created in the above and recognizes the risk level included in the external environment. It is characterized in.

  According to the present invention, robustness with respect to various external environments can be improved when autonomously learning experiences in an actual environment to recognize a risk level.

Basic configuration of risk recognition system Explanatory drawing showing an image area for feature extraction Conceptual diagram of learning by one-dimensional SOM Explanatory diagram showing inefficient part of SOM Explanatory drawing showing hierarchization by division of SOM model Explanatory diagram showing parallelization of SOM model Explanatory drawing showing unit layout by SOM model change Conceptual diagram of state recognition Illustration of pre-learning and online learning Explanatory diagram showing the distribution of self-organizing maps after learning Explanatory diagram showing the relationship between risk level and risk probability Explanatory diagram of risk propagation Illustration of information propagation Explanatory diagram showing expansion of risk information Explanatory drawing showing an output example of recognition results Explanatory diagram showing learning results of risk probability

Embodiments of the present invention will be described below with reference to the drawings.
The online risk learning system of the present invention is a system that is mounted on a moving body such as an automobile and adaptively recognizes information related to the risk (risk) included in the environment from the detection result of the external environment. Will grow online so that it can adaptively recognize risks even in environments that were not envisioned.

  The sensing of the outside environment can use sensing information from various sensor devices provided as an input system of the system, for example, a camera that picks up the outside world with monocular or stereo vision, or a radar device such as a laser or a millimeter wave. That is, this system is not basically dependent on the sensor device that detects the external environment, but in a broad sense, is a learning system that learns the correlation between the external environment information obtained from the sensor device and the risk information.

  In this embodiment, an online risk learning system is applied to a vehicle such as an automobile, and risk information is extracted using image information obtained by imaging the outside world with an in-vehicle camera and vehicle information such as a driver's driving operation and a driving state of the vehicle. An example will be described. In other words, the online risk learning system of this embodiment recognizes the risk by directly associating the risk with the information obtained from the image, and learns the association from the environment encountered in the actual driving, and adaptively Perform risk recognition.

  Specifically, the driver is a teacher in learning of the recognizer, risk information is extracted from the driving operation of the driver, and artificial intelligence technology is used to relate the risk information based on the driving operation and the image information obtained from the camera. To learn. For example, when the driver performs an operation action that avoids a pedestrian, the system determines that the situation is dangerous, and teaches that the image obtained at that time is dangerous.

  As a result, when a similar situation (image) enters the system at the next opportunity, it is output that it is dangerous, and a warning can be given to the driver. In this system, risk is handled probabilistically. As a result, it is possible to handle a case where the risk is different even in a similar situation or an inherently probabilistic risk that cannot be determined only by the obtained image information.

  Hereinafter, the online risk learning system of this embodiment will be described with reference to FIG. The online risk learning system 1 in this embodiment is configured by a single computer system or a plurality of computer systems connected via a network or the like. As a sensor for detecting the external environment of a vehicle, an in-vehicle camera having an image sensor such as a CCD or a CMOS is used.

  Specifically, the online risk learning system 1 includes an image input unit 2, a vehicle data input unit 3, a feature amount extraction unit 4, a state recognition unit 5, a risk information extraction unit 6, a risk recognition unit 7, and a risk information output unit 8. Is provided as a basic configuration. Although the details will be described later, this system performs clustering by converting the feature value mainly extracted from the image captured by the in-vehicle camera into a quantity called a state, and learns the risk from the risk distribution obtained by learning the probability density distribution of the risk for each state. I try to recognize it.

  In the present embodiment, each unit of the SOM is recognized as a state by using a self-organizing map (SOM) which is a kind of hierarchical neural network, and the recognition process is learned. Therefore, the state recognition unit 5 includes a recognition model setting unit 5a that sets an SOM model as a recognition model, and a recognition learning unit 5b that performs recognition learning using the SOM model.

  The general function of each part is as follows. In FIG. 1, illustration of an in-vehicle camera as a sensor for detecting the external environment is omitted.

  The image input unit 2 inputs a captured image from the in-vehicle camera, performs video process processing such as noise removal, gain adjustment, and γ correction, and converts the analog captured image processed by the video process into a digital image with a predetermined gradation. Convert. The image processed by the image input unit 2 is temporarily stored in the memory, collected, and sent to the feature amount extraction unit 4.

  The vehicle data input unit 3 inputs vehicle data at a predetermined cycle from an in-vehicle network or the like. The vehicle data includes vehicle speed, steering wheel angle, accelerator opening, brake pressure, and the like, and these data are sampled at a predetermined cycle (for example, 30 Hz).

  If necessary for subsequent recognition processing, other information flowing on the in-vehicle network, for example, vehicle motion information such as longitudinal and lateral acceleration and yaw rate, system operation information such as headlights and turn signals, outside temperature, and the like are also input.

  The feature amount extraction unit 4 receives the image data from the image input unit 2 and extracts the feature amount of the obtained image. That is, feature amounts such as edge information, motion information, and color information are extracted from the obtained image, and the information is held as an N-dimensional vector. The N-dimensional vector includes vehicle information other than the image feature amount, for example, information such as a change in vehicle speed and yaw angle, as indicated by a broken line in FIG.

  The image data handled in this embodiment is an image captured by a monocular color camera, but may be an image obtained from an infrared camera or a distance image obtained from a stereo camera. In addition, as described above, information from a laser, a millimeter wave, or the like can be used. In this case, the image feature amount should be more generally called an external environment feature amount.

  The state recognizing unit 5 finally converts the obtained N-dimensional feature quantity vector into an amount called a one-dimensional state using the SOM model. The state is an image in which the input image is divided into scenes according to the location where the image is traveling, the weather, the traveling state, and the like. Actually, at the time of online learning, it is not possible to explicitly teach which scene it is now, so the input data is clustered into a class with M states.

  The risk information extraction unit 6 extracts risk information from the input vehicle information (driver operation information), and creates teacher information (risk magnitude and type). In this system, learning about this risk information extraction is not performed, but risk information is recognized using a preset rule.

  The risk recognition unit 7 obtains a correlation between the state obtained by the state recognition unit 5 and the teacher created by the risk information extraction unit 6, and learns and recognizes the risk of the state.

  The risk information output unit 8 outputs the recognized risk by a monitor or voice. Regarding the output of the risk information, in addition to outputting the recognized risk itself, a difference from the risk information obtained from the operation data may be output. For example, even when the risk recognizing unit 7 determines that the risk is high, an output indicating that the risk is not high may be performed if the driver is sufficiently careful by recognizing the operation data.

  In the above online risk learning system 1, the current risk recognition / output and learning are executed in parallel. That is, risk recognition is performed on the input image based on the recognition result learned before that, and the risk learning is performed by updating (learning) the risk recognition unit 7 using the input teacher information. At this time, risk recognition and learning are simultaneously performed by repeating the process of recognizing the risk by the updated risk recognition unit 7.

Learning is possible in each of four parts as shown in (a), (b), (c), and (d) below, but (a) and (b) are large-scale computations for learning. In this system, learning is not performed, and offline learning beforehand or a previously created algorithm is used.
(A) Extraction of risk information from driver operation data (risk information extraction unit 6)
(B) Determination / selection of image feature quantity from image data (feature quantity extraction unit 4)
(C) Conversion from image feature amount to state (state recognition unit 5)
(D) Correlation recognition between state and risk (risk recognition unit 7)

  Next, details of each process in the online risk learning system 1 will be described. In the following, a process for extracting feature values from an image by an in-vehicle camera, a process for recognizing a state from the extracted feature values, a process for extracting a risk from vehicle operation information (creating teacher information), a state and given teacher information Will be described in the order of risk recognition processing.

[Image feature extraction processing]
The feature amount extraction unit 4 extracts data for subsequent risk recognition. In general, data that is not correlated with risk perception adversely affects recognition. That is, in this feature quantity extraction process, it is not a good idea to increase the feature quantity unnecessarily, and conversely, not using a necessary feature quantity also deteriorates accuracy.

  For this reason, feature quantity selection as to which feature quantity should be used occurs as an issue, but as described above, when feature quantity selection is obtained in a learning manner, higher-level learning of risk recognition described below is performed. This is disadvantageous for online learning in terms of computational complexity and memory capacity.

  Therefore, in this embodiment, an example will be described in which the feature amount extraction portion is treated as fixed. When learning, it is only necessary to evaluate the recognition rate of the system as a standard and optimize the combination of each feature quantity. This includes heuristics such as full search of combinations and genetic algorithm (GA). Existing optimization methods such as simple search methods can be used.

  In the present embodiment, a feature amount of a preset type is extracted by the feature amount extraction unit 4. Here, the process is divided into three elements, and feature amounts set for each element are extracted. The three elements are preprocessing, feature amount calculation, and region setting. Specifically, as shown below, 6 types of data are extracted in the pre-processing, 10 types in the feature amount calculation, and 4 types in the region setting, and a total of 240 (6 × 10 × 4) dimensions are obtained by combining them. Extract data.

<Pretreatment>
Six types of filter processing are performed on the input image, sobel, vertical sobel, horizontal sobel, inter-frame difference, luminance, and saturation to extract six-dimensional feature data.

<Feature amount>
Ten types of calculation processing are performed on the pixel values of the filtered image: average, variance, maximum value, minimum value, horizontal centroid, vertical centroid, contrast, uniformity, entropy, and fractal dimension. Feature quantity data is extracted.

<Area>
As shown in FIG. 2, an area A0 is set in the image, and four types of areas, that is, the entire setting area A0, the left area A1, the right area A2, and the central area A3 in the setting area A0 are set to 4 types. Extract dimension feature data.

  Note that the 240-dimensional feature value may be narrowed down according to the computing performance of the online system. For example, in addition to images, vehicle data may also be used to extract 6-dimensional feature quantities such as the average and variance of the Sobel over the entire screen, the average of inter-frame differences over the entire screen, the variance, the vehicle speed, and the steering wheel angle. .

  In the above feature quantity extraction process, each feature quantity is normalized, but the theoretical range is inefficient. Therefore, the distribution of each feature quantity is evaluated in advance, and the evaluation result is used as a basis. The maximum value and the minimum value are set to, and normalized to a numerical value of 0 to 1. In that case, the maximum and minimum values may be changed dynamically. For example, when a value exceeding the maximum value or a value below the minimum value is input, the maximum value is expanded so that the range is expanded.・ Change the minimum value. Conversely, if the data near the minimum and maximum values has not been entered for a while, the range is changed to narrow.

  Although basic feature values are used here, it is also possible to calculate time-series fluctuations of feature values, such as calculating motion information using past frame information, and use the information as feature values. . Furthermore, in order to improve the accuracy of risk recognition as a whole, high-accuracy image processing can be added to this feature amount extraction processing, for example, pedestrian recognition results, road white line recognition results, obstacle recognition results, etc. May be incorporated in the extracted data here. In this sense, the present system can also be regarded as a system that recognizes risks by integrating individual external recognition results.

[Status recognition processing]
The state recognition unit 5 compresses and converts the N-dimensional feature data into an amount called a one-dimensional state. In other words, the state is recognized by the function as a discriminator that outputs the quantity of the state from the input feature data (however, the output of this discriminator can be handled stochastically without determining one state. it can). In this process, the internal structure of this discriminator is adapted to the actual environment using input data and teacher data. However, the teacher here does not directly teach the state of this input data. The risk recognized from the output state can be recognized as efficiently and accurately as possible.

  Specifically, the state recognition unit 5 first sets an SOM model by the recognition model setting unit 5a, and performs recognition and learning by the recognition learning unit 5b using the SOM model. In the SOM algorithm, units arranged in the M dimension (usually two dimensions) each have a vector value (referred to as the weight of the connection with the normal input), and the winner unit with respect to the input is based on the vector distance. decide. Then, the reference vector values of the winner unit and the surrounding units are updated so as to approach the input vector. This is repeated, and learning is performed without a teacher so that the entire distribution of input data can be optimally expressed. An image of learning by one-dimensional SOM is shown in FIG.

  In this case, the SOM unit arrangement (and the probability density distribution of each state) is sequentially updated during traveling, enabling recognition suitable for each user's traveling environment and driving tendency. However, when adapting to a real environment, there is a risk that the robustness may deteriorate in the following situation.

(A) There is a possibility that the risk output has the same value even in a place where the risk of the scene is considered to be greatly different, such as the main road and the back alley. For this reason, there exists a possibility that the robustness with respect to the difference in a place may fall.

(B) The recognition output may be greatly affected by the difference between day and night. For this reason, there exists a possibility that the robustness with respect to the difference in a time slot | zone may fall.

  The decrease in robustness with respect to the difference in the location of (A) is caused by the fact that the learning frequency is uneven among the SOM units of the recognition model, and efficient learning and recognition cannot be performed. In particular, when an SOM unit is arranged on a loop so that there is no end on one dimension, and vehicle information such as vehicle speed and rudder angle and image information are input to the SOM on this one-dimensional loop. Thus, different input distribution regions are being expressed on a one-dimensional loop, such as when traveling and when stopped. For this reason, as shown in FIG. 4, SOM units are arranged even in a sparse portion of input data, which is inefficient in approximating the external environment.

  Therefore, in the present embodiment, as shown in FIG. 5, the SOM of the conventional model on the one-dimensional upper loop is divided into models on a plurality of loops, and the processing is hierarchized using the divided new models. To do. That is, as a first stage process, a model to be used is selected from among models on a plurality of loops, and the same data in the distribution region is input to the selected model. Avoid placing SOM units.

  Further, the decrease in robustness with respect to the difference in time zone (B) is caused by the fact that the probability density distribution itself is averaged when learning is performed with data having different properties such as daytime and nighttime. Learning that targets only one of the data with different properties cannot handle the other environment. Therefore, as a higher level of model hierarchization to deal with the situation of (A), for example, a plurality of model groups such as a day model and a night model as shown in FIG. By switching model groups, the robustness against data input with different properties is improved.

  The following description will be divided into a process for setting an SOM model, a process for recognizing a state using the set SOM model, and a learning process for learning the recognition process itself.

<Setting of SOM model>
The recognition model setting unit 5a first selects a model group to be used in a relatively long time from among the SOM model groups that are parallelized according to the nature of the input data. For example, in the case of having a daytime SOM model group and a nighttime SOM model group, the models are switched according to day and night determined from the lighting state of the small light switch. However, when switching models, hysteresis is provided as a countermeasure against passing lights and the like.

  The parallel model group includes a daytime model and a nighttime model, for example, a rainy weather model and a clear weather model, a general road model, and a highway model. By switching between these model groups, it is possible to expect an improvement in robustness in an environment where the tendency is greatly different and fluctuates on a long time axis of at least several minutes or more.

At this time, as information used for switching the parallel model, for example, the following information (1) to (6) can be used.
(1) Wiper operation information for discrimination in rainy weather (2) Sunlight sensor information for discrimination in fine weather (3) Vehicle speed and gear information for discrimination of highway driving (4) Travel position of own vehicle GPS information for discrimination of time, season, (5) Camera control information such as shutter speed for discrimination of brightness conditions, (6) Image information for each discrimination

  Next, the recognition model setting unit 5a selects an SOM model to be used in a relatively short span from the selected model group according to the traveling state determined from the vehicle information. Specifically, each model group is divided into four models at stopping, right turn, left turn, and normal driving. As a first step, vehicle speed, steering angle, etc. are selected from the four models. Select the model to use according to the information.

  The graph shown in FIG. 7 shows the unit arrangement by changing (separating) the SOM model, where the horizontal axis is the unit number and the vertical axis is the appearance frequency (occurrence frequency) of each unit. In FIG. 7, unit numbers 0-4 are when turning right, unit numbers 5-9 are turning left, unit numbers 10-14 are stopped, and unit numbers 15-99 are during normal driving. For convenience, when turning right, turning left Four models are displayed on a single axis for hour, stop, and normal travel.

  From FIG. 7, it can be seen that there were units that were rarely used in some places before the model change (separation), but the unit was improved by the model change and an efficient unit arrangement. In addition, the unit number 10 vicinity after a change is a unit corresponding to the time of a stop, and it is a special situation that the number of appearance of a unit is many during a stop.

<Recognition process>
The recognition learning unit 5b inputs the same kind of data of only the image feature amount to the selected SOM model as the second stage process for the SOM model selected according to the driving state, and learns the state recognition process and the recognition process. I do. This recognition process is performed as a prototype type identification process for input data. Here, if the state number is S, each state has a representative value, which is designated as prot _s (i). The state representative value prot _s (i) is an N-dimensional vector, i = 0, 1,..., N−1.

Assuming that the input data (feature vector) is In (i), the input vector is obtained by the distance L (s) from the state representative value prot _s (i) as shown in the following equation (1). Whether it belongs to a state is recognized.
L (s) = (Σ _i (prot _s (i) −In (i)) ² ) ^1/2 (1)

The state (state number) K to which the input data belongs is obtained by the minimum value of the distance L (s) as shown in the following equation (2), and is recognized as the state representative state having the closest input vector. Is done.
K = min _s (L (s)) (2)

  FIG. 8 shows a case where attention is paid to three of the N dimensions. Since the input data is closer to the state S1 than the state S6, the input data is recognized as being in the state of S1. The above is the basic state recognition, which is to determine the state of the input data.

  In this case, in FIG. 8, the distance between the state S1 and the state S6 is not so different, but the input data is recognized as the state S1 because the distance to the state S1 is slightly close. That is, in a region where the distance between the state S1 and the state S6 is almost the same, the recognition may become unstable.

Therefore, here, the instability of recognition is resolved by further expanding and treating the state as probabilistic. That is, if the probability that the input data is in the state s is P (s), the probability of the state is obtained by the following equations (3) and (4) using the distance L (s). Here, σ is a parameter, and the smaller the value, the more effective the state becomes deterministic.
P (s) = (exp (−L (s) / σ)) / z (3)
z = Σ _s exp (−L (s) / σ) (4)

  As described above, when states are stochastically determined on a scale according to the distance from the input data, it is necessary to calculate all states in the subsequent calculations. Therefore, in order to reduce the amount of calculation, the probability below a certain value may be set to 0 and may not be treated as a calculation.

  In the definition of P (s), if 1 is set only when s = K and 0 is set otherwise, it is the same as when the state is fixed.

<Learning process>
Next, in the learning process of the state recognizing unit 5, learning (update) of the representative value of each state based on the SOM is performed from the input data and the teacher information. In this system, learning by SOM is as follows. However, in this system, it is assumed that units (states) are connected in one dimension. When the state number of the winner unit is the above-described state number K, the representative vector prot _s is updated (learned) according to the following equation (5).
prot _s (i) → prot _s (i) + α (In (i) −prot _s (i) (5)

Here, α in the equation (5) is a learning rate coefficient indicating the weight of update, and is expressed by the following equation (6).
α = a · b (t) · c (D (s, K), t) · e (t) (6)
Where a: learning coefficient
b: Time decay coefficient
c: Domain attenuation coefficient
D (s, K): Distance in the connection between the unit to be updated and the winner vector
e: Teacher information coefficient

  The parameters a, b, and c in the equation (6) are parameters that are also used in normal SOM, and the time attenuation coefficient b is a function of the learning elapsed time t (usually indicating how many times it is updated). Decreases with increasing time t. Further, the distance D (s, K) is not a distance in the feature amount space. For example, in FIG. 3, the unit adjacent to the winner unit is the distance 1 and the adjacent unit is the distance 2.

  On the other hand, the region attenuation coefficient c is a function of the distance D (s, K), and the value decreases as the distance D (s, K) increases. It is set not to be updated. The region attenuation coefficient c is also a function of the time t, and the value decreases as the time t increases. Furthermore, in this system, a teacher information coefficient e (t) indicating teacher information is introduced, which will be described later.

  As described above, in the learning algorithm of SOM, in the initial stage of learning, a wide range of units are updated so as to approach the input data. As the learning progresses, both the number of units to be updated and the amount of update are reduced. The rate coefficient α (update weight) becomes 0, and learning ends. In the initial state, the units are usually randomly arranged near the center on the vector space.

  Here, since this system is an online learning system, the above learning algorithm is slightly changed, the learning is divided into a pre-learning phase and an online learning phase, and learning parameters are changed in each learning phase. In other words, the learning end time is not set, the time is constant in the pre-learning phase and the online learning phase, the update range is not attenuated, and the range for pre-learning and online learning is different. To do.

  This is because, in this system, which is online learning, there is no end time for learning, and pre-learning is introduced because it is uniquely set to a constant value. This is because the data is concentrated in a narrow range near the average of the input data. On the other hand, if the input data is small, the SOM distribution is too varied in the feature amount space, and the input data distribution cannot be expressed well.

  Therefore, as shown in FIG. 9, by taking large values of the time attenuation coefficient b and the area attenuation coefficient c as pre-learning, first, the SOM is brought close to the center of the input data distribution, and then the time attenuation coefficient is set. By reducing b and the region attenuation coefficient c, an appropriate distribution can be expressed. The pre-learning here is assumed to be off-line learning before being used for actual driving in the market.

  FIG. 10 shows an example of SOM distribution after learning. Although the actual feature amount space is 240 dimensions, FIG. 10 shows only three dimensions, and each point of the graph indicates input data. Actually, each point is expressed as a colored point, and the magnitude of the risk is represented by the color. Black dots are representative vectors for each state, and black lines connecting them are SOM connections.

  In the above, learning methods that can optimally express the distribution of input data have been described, but what is actually required is that the distribution of input data can be optimally expressed in order to recognize risks. SOM is originally an unsupervised learning method (incoming data is treated equally and learned), but in this system, the above-mentioned teacher information is used as an efficient learning method for recognizing risks. Learning is performed with risk information given by a coefficient e (t).

  Although details will be described later, the risk recognition is output in the form of the risk probability of the recognized state. This represents the probability that the state has a risk. As a specific learning method, when input data at time t has teacher information called risk level R obtained from driver information, if the risk probability of the recognized state has a high probability of risk level R, the teacher information The coefficient e (t) is increased, and if it is smaller, the teacher information coefficient e (t) is decreased. If teacher information cannot be obtained, the teacher information coefficient e (t) is reduced.

  As a result, as the learning progresses, the recognized state has the risk at that time with a high probability, that is, the risk recognition accuracy is improved. The specific setting of the teacher information coefficient e (t) will be described in the next risk recognition process.

  For learning when the state is obtained stochastically, the winner unit is expressed by a weight corresponding to the probability, and the update is performed with an update amount corresponding to the weight. However, since there is a problem that the calculation amount increases, in this system, the learning is performed with the winner unit determined to be closest to the input data at the time of learning. Does not update itself as a winner.

[Risk information extraction process]
As described above, in this embodiment, when extracting risk from vehicle operation information, the risk information extraction unit 6 does not perform learning, but extracts risk information from driver operation information using a preset rule. . In the risk information extraction process according to this rule, the risk information is handled as one-dimensional data with a level.

  Specifically, the risk level is set to 11 levels from 0 to 10 (integer value), and the higher the value, the higher the risk. However, the risk information here is not intended to recognize the risk at regular intervals, such as every 30 Hz frame. This is because the actual driver's operation is not performed only by risk, and it is considered that the proportion of operations that accompanies risk is a part that is less than 10% of the total driving, for example. is there.

  That is, the first goal is to recognize risk information from driver operation data only when there is a certain degree of risk that affects the driver's operation behavior. Therefore, for risk 0, the output indicates not only that there is no risk, but also that there is no teacher information.

  In addition, risk recognition rules are arbitrarily set so as to fit the reality as much as possible. The rules shown in (1) to (5) below are parallelized, and the risk with the largest value in each condition is selected. Teacher risk.

(1) Did you brake suddenly? Set the risk level according to the difference in brake pressure between frames. For example, if the difference in brake pressure is 1 × 10 ² kPa or more, there is a risk, 1 × 10 ² kPa is risk 5, 1 × 10 ³ kPa is risk 10, and the difference between the brake pressure is between risk 5 and risk 10. Set linearly according to.

(2) Did you step on the brake? Set the risk level according to the brake pressure at a specified vehicle speed or higher. For example, when the vehicle speed is 10 km / h or higher and the brake pressure is 20 × 10 ² kPa or higher, the risk is 10, when the brake pressure is 10 × 10 ² kPa or higher, the risk is 6, and when the brake pressure is 5 × 10 ² kPa or higher. Is risk 2.

(3) Has the steering wheel been turned off? If the absolute value of the difference in the steering wheel angle between frames is greater than or equal to a set value (for example, 10 degrees) with no turn signal, risk 5 is assumed.

(4) Was the accelerator suddenly released? The risk level is set according to the difference in accelerator opening between frames at a predetermined vehicle speed or higher. For example, if the vehicle speed is 5 km / h or more and the difference in accelerator opening is -1% or less, the risk is 4.

(5) Is the accelerator stepped on? Set the risk level according to the accelerator opening during acceleration. Whether or not the vehicle is accelerating is determined by the differential value of the vehicle speed. If the differential value of the vehicle speed is 0 or more (during acceleration) and the accelerator opening is 1% or less, the risk is 2.

  Of course, the above rules can be added and deleted, and can be adjusted to be more realistic. It is also possible to perform “learning acquisition of risk recognition from driver data” by incorporating an algorithm for automatically generating the above rules, and further incorporating fuzzy elements into the above rules.

[Risk recognition process]
The risk recognition unit 7 outputs a risk according to the state obtained by the state recognition unit 5. As described above, since each state has a risk probability distribution, the risk probability distribution in state s is represented as p (R | s). The risk here corresponds to the risk in the risk information extraction unit 6 and is divided into 11 levels, so the risk level R and the risk probability (distribution) p (R | s) are: For example, it is represented by the relationship shown in FIG.

The risk output basically outputs this risk probability p (R | s). However, when the output result is used for alarm or display, for example, it is difficult to use the probability distribution as it is. Outputs an expected value E expressed by the following equation (7).
E = Σ _R R · p (R | s) (7)

Further, when the state is handled stochastically, the expected value E is expressed by the following equation (8).
_{E = Σ s Σ R P (} s) · R · p (R│s) ... (8)

<Risk probability learning process>
The learning of the risk probability is performed every frame, and the risk probability is updated sequentially. The risk probability is basically calculated using a frequency distribution of risk levels experienced in the past. However, because this system is online learning, it is difficult to have past data at infinity, and it is not desirable to have distant past experience as important as the present. Therefore, the risk probability is updated here by the following method.

When the risk probability of state s _t at time t and the p _t (R￨s _t), in accordance with the following equation (9), and updates the risk probability.
p _{t + 1} (R | s _t ) = p _t (R | s _t ) + β (9)

Further, the risk probability p _{t + 1} (R | s _t ) is normalized according to the following equation (10).
p _{t + 1} (R | s _t ) ← p _{t + 1} ( _R | s _t ) / ΣR p _{t + 1} ( _R | s _t ) (10)

Note that the status update is only the status at that time. When the states are handled stochastically, β is calculated as p (s _t ) · β in each state. Here, β is a constant, and the larger this value, the more important the current information is.

  Here, as described in the explanation of the risk information extraction unit 6, the given teacher risk is not necessarily obtained for each frame. When the risk level is high, risk information is often obtained from the driver data. However, when the risk level is low, there is a problem that the possibility of obtaining teacher information is particularly small.

  This system addresses this problem by propagating teacher risk information in the time axis direction. This is based on the causal relationship that if the teacher risk information is obtained at a certain time, it is dangerous even if the previous time is not the same as that time. To propagate.

  In this case, in order to propagate information in the past, it is necessary to memorize all past state transitions for the amount to be propagated. Become. Therefore, in this system, the risk probability is updated by propagation in consideration of a TD (Temporal Difference) error used in reinforcement learning.

  Reinforcement learning is not based on an explicit action instruction for the current state, but learning based on the reward for the action performed, and the action that maximizes the sum of the rewards that can be obtained in the future is It is a learning method that is selected from time to time, and the difference between the actual reward and the predicted value of the reward at time t is called a TD error (TD-ERROR), and learning is performed so that this is zero. The risk information of this system corresponds to the reward of this reinforcement learning. As shown in FIG. 12, when considering the state transition in a certain scene, the states S2, S7,... Propagate risk information.

  In this case, propagation only needs to be propagated from the current state to the previous frame (that is, the calculation and storage need only be handled in relation to the previous frame), and the risk information is sufficient in one experience. Although it does not propagate to the past, by repeating the same experience, risk information gradually propagates and the causal relationship can be learned. Further, as shown in FIG. 13, the propagation of the risk information is not only propagated from the state St at the time t at the same risk level to the state St-1 at the time t-1, but also between the states at different risk levels. I will let you. However, risk level 0 includes no risk information and no risk information, and therefore does not propagate.

The update of the risk probability p (r | s _t-1 ) due to propagation is performed by the following equation (11).
p (r | s _t-1 ) = p (r | s _t-1 ) + η · (RI (r) + γ · p (r | s _t ) −p (r | s _t-1 ))
+ H · η · (γ · p (r−1 | s _t ) −p (r ₋₁ | s _t−1 ))
+ H · η · (γ · p (r + 1 | s _t ) −p (r + ₁ | s _t−1 )) (11)
Where h is a parameter indicating the magnitude of propagation in the risk level direction
γ: Parameter indicating the magnitude of time series propagation
η: A parameter that represents the size of the update in a single learning

Here, the risk information obtained at time t is expressed as RI (r) using the risk level r. As described above, the risk information handled by the risk information extraction unit 6 is assumed to be obtained for a certain risk level among 11 stages of 0 to 10. That is, if the risk information obtained at time t is the risk level Q, it is expressed as in equations (12) and (13).
RI (r) = 1 (r = Q) (12)
RI (r) = 0 (r ≠ Q) (13)

  On the other hand, as shown in FIG. 14, the risk information RI (r) in this risk learning is expanded on the assumption that there is actually a risk near the risk level. More specifically, the expansion is performed by multiplying the adjacent risk level by g using the parameter g (g <1) and multiplying the adjacent risk level by g * g. There is an effect that limited teacher data can be used more effectively. Further, h representing the magnitude of propagation in the risk level direction is normally set to the same value as g used for expanding risk information.

After updating the risk probability, the teacher information coefficient coefficient e (t) used in the learning process in the state recognition unit 5 is set. The teacher information coefficient e (t) is set according to the following equations (14) and (15). The risk information obtained from the risk information extraction unit 6 at time t is used as R,
When R ≠ 0
e (t) = 10 · R · p (R | s _t ) (14)
When R = 0
e (t) = const (15)

  When R = 0, this corresponds to the case where teacher information is not entered. In this case, the teacher information coefficient e (t) is a constant const, that is, a fixed value gain. This value is determined by the probability that a teacher risk can be obtained, and is set based on the ratio between the number of learning data with teacher and the number of learning data without teacher. In this system, as a rule of thumb, a constant const is set so that the number of learning data with teacher = the number of learning data without teacher is set, and const = 0.01.

  When teacher information is entered, the higher the probability and the higher the risk level, the stronger the learning. Thereby, when the probability of recognizing an event that has actually occurred is low, it indicates that there is a high possibility that the recognition of the state is wrong, and learning is weakened. The representative vector in that state is learned so that it recognizes the same state and approaches data with a high probability of risk. As such learning continues, the input data is easily recognized as another state, and it is difficult to recognize a state that is likely to be wrong. In this way, state recognition and risk recognition as a whole are optimized.

  An output example of the recognition result by the above processing is shown in FIG. FIGS. 15A to 15D are output images of a system in which the recognition result is displayed on the image obtained from the in-vehicle camera, and the magnitude of the recognized risk is represented by bar graphs B1 to B4 at the bottom of each screen. It is represented by The recognition risks represented by the bar graphs B1 to B4 indicate the expected value of the risk probability described above, and the numbers displayed thereon are the recognized state numbers.

  The two images shown in FIGS. 15 (a) and 15 (b) are scenes in which pedestrians and oncoming vehicles are not nearby and are considered to be low in risk, and FIGS. 15 (c) and 15 (d). The two images shown in Fig. 15 are a scene in which an oncoming vehicle is present on a one-lane road with a narrow road width and the road width is further reduced, and a left-turn scene at an intersection. , (B) is a scene that seems to have a higher risk.

Here, “the risk is considered to be low (high)” is described because there is no absolute value that the number of risk values for each image. From the recognition results of this system, it can be seen that the scenes of FIGS. 15C and 15D are recognized as having a higher risk than the scenes of FIGS. 15A and 15B. FIG. 16 shows the learning result of the risk probability p _t (R | s _t ) at this time.

DESCRIPTION OF SYMBOLS 1 Online risk recognition system 4 Feature extraction part 5 State recognition part 5a Recognition model setting part 5b Recognition learning part 6 Risk information extraction part 7 Risk recognition part

Claims (5)

An online risk learning system that detects the external environment of a mobile object and recognizes the risks contained in the external environment in a learning manner.
A feature quantity extraction unit that processes detection information of the outside environment and extracts multidimensional feature quantities included in the outside environment;
A recognition model setting unit for setting a recognition model for clustering the feature quantities in a hierarchical manner;
A state recognition unit for recognizing the multidimensional feature quantity as a one-dimensional state using the recognition model set by the recognition model setting unit;
Based on the correlation between the state recognized by the state recognition unit and the teacher information created by extracting risk information related to the risk, the risk of the state is adaptively learned and included in the external environment An online risk learning system comprising a risk recognition unit for recognizing the degree of risk.   The online system according to claim 1, wherein the recognition model is created by a self-organizing map, and the self-organizing map is divided into a plurality of one-dimensional models according to the driving state of the moving body to be hierarchized. Risk learning system.   One one-dimensional model is selected from the plurality of one-dimensional models according to the driving state of the moving body, and only the feature amount extracted from the image information obtained by imaging the external environment is input to the selected one-dimensional model. The online risk learning system according to claim 2, wherein:   4. The online risk learning system according to claim 2, wherein the plurality of one-dimensional models are parallelized according to the nature of input data.   5. The online risk learning system according to claim 4, wherein the plurality of parallel one-dimensional models are switched with hysteresis.