JP6206022B2 - Driving assistance device - Google Patents

Description

  The present invention relates to a driving support device that provides information according to the situation of a host vehicle.

Conventionally, various information for safely driving according to the situation of the host vehicle, such as the speed and distance of surrounding vehicles, and a margin time for collision, has been presented to the driver.
As one of the technologies for presenting such information, depending on the degree of danger set based on the detection results of various sensors, etc. For example, a technique for visualizing information desired to be presented to the driver in a natural manner by coloring the road is known (for example, see Non-Patent Document 1).

C. Roessing et al., "Intuitive Visualization of Vehicle Distance, Velocity and Risk Potential in Rear-View Camera Applications," IEEE Intelligent Vehicles Symposium, 2013

  The conventional technology determines whether the external situation grasped from the detection result of the sensor or the like is similar to a predefined driving scene (for example, “waiting for a signal”, “sudden stop”, etc.), and On the other hand, information is presented in a pre-defined manner at a pre-defined part. For this reason, the designer needs to clearly define the criteria for determining the driving scene for all driving scenes for which information is to be presented.

  However, since the actual driving scenes are extremely diverse, it is difficult to assume every driving scene from the start to the end of driving, and in reality it is limited to specific driving scenes that can be assumed by the designer. However, there was a problem that only information could be presented.

  Also, even if the external situation (for example, vehicle speed and positional relationship with other vehicles) is the same, depending on the driver's driving intention (for example, whether or not he intends to change lanes), the site where the driver should be alerted varies. It will be. However, the difference in the comprehensive situation (hereinafter also referred to as “driving context”) including the driving intention of the driver cannot be identified only by the instantaneous external situation detected by the sensor or the like. That is, the conventional technology has a problem that only a specific driving context excluding the driving intention can be handled.

  Furthermore, when a certain external situation is detected, it may be possible to present all the information to be presented in each of a plurality of driving contexts that can take that external situation. There was a problem that it was not possible to understand whether the information was really necessary information.

  In order to solve the above-described problems, an object of the present invention is to realize information presentation according to a comprehensive situation including a driving intention of a driver in a driving support device.

The driving support apparatus of the present invention includes driving behavior data collecting means, environmental data collecting means, scene dividing means, feature quantity distribution generating means, and notification means.
The driving behavior data collecting means repeatedly collects driving behavior data related to the driving behavior of the vehicle, and the environmental data collecting means repeatedly collects environmental data representing the environment around the vehicle. The scene dividing means divides the series of driving behavior data collected by the driving behavior data collecting means into a plurality of partial series each representing some kind of driving scene. The feature quantity distribution generation means is a local feature obtained from the driving feature quantity and the environmental data collected by the environmental data collection means, as one or more types of feature quantity obtained from the driving behavior data collected by the driving behavior data collection means. A feature amount distribution is generated that represents the appearance frequency for each index indicating which feature value belongs to which bin the value range of the feature value is divided into a plurality of bins for each partial series and for each feature value, with the amount as an environmental feature value. To do. The notifying unit controls the notification regarding the local region of the environmental data by using the co-occurrence between the driving feature amount distribution and the environmental feature amount distribution in the feature amount distribution group.

  According to the present invention configured as described above, for each driving scene based on the driving intention of the driver obtained by the scene dividing means, the environmental data based on the co-occurrence of the driving feature quantity distribution and the environmental feature quantity distribution In order to obtain the strength of the relationship with the driving behavior data, the notification means can set an appropriate local region of environmental data for each driving scene. Therefore, according to the present invention, it is possible to realize information presentation according to a comprehensive situation including a driver's intention to drive.

  Note that the notification means can be constituted by, for example, a topic storage means, a ratio calculation means, a probability distribution generation means, and a notification control means. In this case, a plurality of types of feature quantity distributions generated by the feature quantity distribution generating means for each partial series are used as a feature quantity distribution group, and the topic storage means is used as a base distribution when expressing the feature quantity distribution group. A driving topic group composed of a plurality of driving topics is stored. In other words, the driving topic is a set of base distributions of driving feature amounts and environmental feature amounts, and the feature amount distribution group generated for each partial series is expressed as a mixture of a plurality of driving topics. At this time, the mixing ratio of this driving topic is called a topic ratio.

  The ratio calculation means calculates a topic ratio, which is a mixing ratio when the feature quantity distribution group is expressed by the driving topic group, for each partial series. The probability distribution generation means probabilistically indicates to which driving topic the local region having the environmental feature amount is derived for each environmental feature amount based on the topic ratio and the driving topic group for each partial series. Generate an allocation probability distribution. The notification control means assigns a driving topic for each local region according to the environmental feature amount and the allocation probability distribution, and controls notification regarding the local region to which the driving topic satisfying a preset presentation condition is assigned.

  The individual driving topics that make up the driving topic group represent specific situations that repeatedly appear during driving, and the feature quantity distribution group in each subseries is expressed by individual driving topics or combinations of driving topics. The In the future, information on driving that each sub-series has will be called driving context. That is, in the present invention, the driving context is expressed by the driving topic and the topic ratio.

  In the present invention, a series of driving behavior data is divided into partial series each representing some driving scene, and each partial series is grasped from a feature amount distribution generated based on driving behavior data and environmental data. A driving context is extracted in the form of a driving topic and a topic ratio, a local area associated with the extracted driving context is estimated, and notification regarding the estimated local area is performed.

  Thus, according to the present invention, instead of determining which of the predefined driving scenes the current situation is similar to, what driving topic the extracted driving scene is composed of The driving context is extracted by analyzing the situation. For this reason, in the present invention, unlike conventional devices, information provision according to the driving context is realized not only for the limited driving scenes defined in advance but also for all driving scenes that appear from the start of driving to the end of driving. can do.

  In addition, the code | symbol in the parenthesis described in the claim shows the correspondence with the specific means as described in embodiment mentioned later as one aspect, Comprising: The technical scope of this invention is limited is not.

  In addition to the above-described driving support device, the present invention includes various modes such as various systems including the driving support device, programs for causing a computer to function as each means constituting the driving support device, and a driving support method. Can be realized.

It is a whole block diagram of a driving assistance device. It is a histogram which illustrates feature-value distribution. It is a graphical model assumed by Latent Dirichlet Allocation (LDA). It is explanatory drawing which shows the production | generation process of a driving | running topic, and the calculation process of the topic ratio using a driving | running topic. It is explanatory drawing which illustrated the base feature distribution regarding driving behavior data. (A) is an explanatory diagram showing a process of generating an allocation probability distribution from a topic ratio, and (b) is an explanatory diagram showing a process of allocating a driving topic for each local region according to the allocation probability distribution. It is explanatory drawing which illustrated the feature emphasis distribution which converted the appearance frequency in a base feature distribution into likelihood ratio. It is a figure which illustrates the emphasis display according to a driving context, (a) shows the case where "high brake topic" (b) is designated as a high-speed driving | running topic as a remarkable topic used as the object of emphasis display. . It is a figure which illustrates the emphasis display according to driving context, (a) shows the case where distance image data is used, (b) shows the case where image data is used.

Embodiments of the present invention will be described below with reference to the drawings.
<Overall configuration>
As shown in FIG. 1, the driving support device 1 includes a driving behavior data collection unit 2, an environment data collection unit 9, a driving scene discretization unit 3, a feature amount distribution generation unit 4, a driving topic database 5, A topic ratio calculation unit 6, an allocation probability calculation unit 7, and an enhancement region setting unit 8 are provided.

  The driving support device 1 is realized by a microcomputer (microcomputer), and each unit except the driving topic database 5 is realized by a CPU (not shown) included in the microcomputer executing a predetermined program. That is, FIG. 1 shows various functions realized by the CPU divided into functional blocks. However, these units do not necessarily have to be realized by software, and all or a part thereof may be realized by hardware such as a logic circuit.

<Driving behavior data collection unit>
The driving behavior data collection unit 2 repeatedly collects driving operation data related to driving operations by the driver and vehicle behavior data related to vehicle behavior appearing as a result of the driving operations via various sensors mounted on the vehicle, Differential data (dynamic features) obtained by differentiating driving operation data and vehicle behavior data is generated, and multidimensional data including the driving operation data, vehicle behavior data, and differential data is output as driving behavior data.

  As the driving operation data, for example, an accelerator pedal operation amount, a brake pedal operation amount, a steering wheel operation amount (steering angle), a direction indicator operation state, a transmission shift position, and the like can be used. As the vehicle behavior data, for example, vehicle speed, yaw rate, or the like can be used. Hereinafter, individual data constituting the driving behavior data is also referred to as individual data.

<Environmental Data Collection Department>
The environmental data collection unit 9 repeatedly acquires image data output from the in-vehicle camera 11 installed so as to photograph the vehicle periphery such as the front, side, and rear of the vehicle, and outputs the image data to the feature amount distribution generation unit 4 To do.

<Operation scene discretization unit>
The driving scene discretization unit 3 statistically analyzes the driving behavior data obtained from the driving behavior data collection unit 2 using a model from the driver's environmental recognition to the operation, and the driving felt by the individual driver (or a general driver). By extracting scene switching points, the time series of driving behavior data is segmented (discretized) into a plurality of partial series each representing some driving scene.

Specifically, a double articulation analyzer (DAA) that performs segmentation by an unsupervised driving scene division method using a double segment structure is used.
DAA first defines in advance a cluster representing various vehicle states grasped from driving behavior data in a multidimensional space representing a range of driving behavior data, and transition probabilities between the clusters. By using information to statistically determine which cluster the driving behavior data acquired from the driving behavior data collection unit 2 belongs to, the time series of the driving behavior data is obtained for each vehicle state ( That is, it is classified into each cluster). However, the time series of the driving behavior data is converted into a symbol string indicating which cluster belongs by associating each cluster with a symbol for identification in advance. For the generation of this symbol string, for example, a Hierarchical Dirichlet Process Hidden Markov Model (HDP-HMM) which is one of models expressed by a hidden state and a stochastic transition between the states is used. can do. However, in HDP-HMM, the clusters and the transition probabilities between the clusters are generated in advance by learning.

  Next, DAA uses a Nested Pitman-Yor Language Model (NPYLM), which is an example of an unsupervised chunking technique for discrete character strings using statistical information, to generate a predetermined operation scene. Segment into meaningful subsequences. At this time, the generation probability of the entire symbol string composed of the partial series is maximized. This makes it possible to segment the driving behavior data into driving scenes. However, transition probabilities between partial sequences and partial sequence generation probabilities are generated in advance by learning.

  As for DAA to which HDP-HMM and NPYLM are applied, non-patent literature, T. Taniguchi et al, "Semiotic Prediction of Driving Behavior using Unsupervised Double Articulation Analyzer," IEEE Intelligent Vehicles Symposium, 2012 , And K. Takenaka et al, “Contextual Scene Segmentation of Driving Behavior based on Double Articulation Analyzer,” IEEE / RSJ International Conference on Intelligent Robots and Systems, 2012, etc. To do. However, the method used for symbol generation and symbol segmentation is not limited to HDP-HMM or NYLM, and other methods may be used.

<Feature distribution generation unit>
The feature quantity distribution generation unit 4 includes a driving feature calculation unit 41 and an image feature calculation unit 42.
The driving feature calculation unit 41 is based on the time-series driving behavior data output from the driving behavior data collection unit 2 for each partial series (that is, any driving scene) segmented by the driving scene discretization unit 3. Then, a distribution of feature quantities appearing in the partial series is generated.

  In the present embodiment, eight types of individual data constituting the driving behavior data (accelerator operation amount (accel), brake pedal operation amount (brake), steering operation amount (steering), vehicle speed (velocity), and these Of the differential data Δaccel, Δbrake, Δsteering, Δvelocity) is generated as a feature amount.

  As shown in FIG. 2, the feature quantity distribution is obtained by dividing a feature space representing a value range of the feature quantity into a plurality of parts (in the figure, the minimum value to the maximum value of the feature quantity are divided into 20 equal parts), and bin indexes. Each is composed of a histogram expressing the appearance frequency of the feature amount. In the figure, (a) shows the vehicle speed V, and (b) shows the characteristic amount distribution of the differential value (acceleration) ΔV of the vehicle speed.

  The image feature calculation unit 42 segments the image data supplied from the environment data collection unit 9 according to the partial series segmented by the driving scene discretization unit 3, and the distribution of the feature amount appearing in the partial series of the image data Is generated. Here, the input image is compressed to a predetermined size (for example, 320 × 240 pixels), and the compressed image data is compressed into a plurality (for example, 24 × 24 pixels) having a predetermined size (for example, 24 × 24 pixels) at regular intervals (for example, 5 pixels). Here, 3072 points) local regions are set, and SIFT (Scale-Invariant Feature Transform) feature values obtained for the respective local regions are set as feature amount distribution generation targets. The index of the feature quantity distribution is determined in advance by learning. For example, the feature space representing the SIFT feature value range is clustered using a clustering technique such as k-means, and each cluster obtained as a result is used as an index of the feature value distribution. However, the feature quantity for generating the distribution from the image data is not limited to the SIFT feature quantity, and other local feature quantity extraction methods may be used.

  Hereinafter, the feature quantity distribution generated by the driving feature calculation unit 41 is also called a driving feature quantity distribution, and the feature quantity distribution generated by the image feature calculation unit is also called an image feature quantity distribution. That is, the feature quantity distribution generation unit 4 is a feature quantity distribution group including a driving feature quantity distribution for each individual data (here, 8 pieces) and one image feature quantity distribution for each partial series corresponding to the driving scene. Are output together with the image feature amount (SIFT feature amount) used to generate the image feature amount distribution.

<Operation topic database>
The driving topic database 5 stores in advance a plurality (for example, 100) of driving topics used as a basis when the feature amount distribution group generated by the feature amount distribution generation unit 4 is expressed by a mixture of a plurality of distributions. Has been.

Here, a method for generating a driving topic will be described.
For the generation of driving topics, each partial series corresponding to one of the driving scenes is regarded as “one document”, and observed features (such as driving feature amounts and SIFT feature amounts) are regarded as “one word”. Then, the topic estimation method used in the natural language processing field is used.

  In particular, in order to estimate driving topics across multiple modalities consisting of driving behavior data (driving feature quantity distribution) and image data (image feature quantity distribution), Latent Dirichlet Allocation (LDA) is used. A multimodal LDA, which is an extended method, is used.

In LDA, a graphical model as shown in FIG. 3 is assumed. However, D is the total number of scenes, Md is the total number of frames in the d-th scene, K is the total number of topics, w _{d, m} are the features observed in the m-th frame of the d-th scene (ie, driving feature amounts), z _{d, m} feature w _d, (taking natural number 1 to k) topic instruction variable that instructs the topics assigned to _m, multinomial distribution parameter indicating the percentage of topics theta _d is included in the d-th scene, phi _k Is a multinomial distribution parameter indicating the rate at which each feature quantity is generated from the kth topic, and α and β are parameters of θ and φ, respectively. Further, θ _d , φ _k , z _{d, m} w _{d, m} are defined by the equations (1) to (4), where Dir is a Dirichlet distribution and Multi is a multinomial distribution.

Θ and φ are estimated using this model, and approximation methods such as variational Bayes and Gibbs sampling can be used for these estimations. For details on these methods, see, for example, D. Blei et al, "Latent Dirichlet Allocation," Journal of Machine Learning Research, 2003, and T. Griffiths & M. Steyvers, "Finding Scientific Topics," Proceedings of the National Since it is described in Academy of Sciences, 2004, etc., explanation is omitted here. The extension to multimodal LDA is also described in T. Nakamura, et al, "Grounding of word meanings in multimodal concepts using LDA," IEEE / RSJ International Conference on Intelligent Robots and Systems, 2009, etc. Then, explanation is omitted.

However, the learning feature amount distribution (driving feature amount distribution, image feature amount distribution) used when generating the driving topic is, for example, generated as follows.
First, by actually running the vehicle and taking an image for learning with the in-vehicle camera 11, by operating the driving behavior data collection unit 2, the driving scene discretization unit 3, and the feature amount distribution generation unit 4, A learning feature amount distribution corresponding to the learning image is generated. Next, a result obtained by estimating a distribution serving as a basis by multimodal LDA using the generated feature distribution for learning is a driving topic.

  In addition, as shown in FIG. 4, the driving topic includes a plurality of driving base distributions corresponding to individual individual data (driving feature amount distribution) and an image base distribution corresponding to image data (image feature amount distribution). It is configured. Hereinafter, when the driving preference distribution and the image base distribution are not distinguished, they are collectively referred to as a base feature distribution.

  FIG. 5 illustrates the driving base distribution for five driving topics TP1 to TP5 among the generated driving topics. The appearance frequency for each feature index is represented by a tone, and the whiter the tone, the higher the appearance frequency.

  Each driving topic constituted by the driving base distribution generated in this way represents a specific situation that repeatedly appears during driving. For example, the driving topic TP1 appears in a situation where the steering operation is frequently performed in a low speed region as compared with other driving topics. Then, some driving context is expressed by these individual driving topics or combinations of driving topics.

<Topic ratio calculation unit>
In the topic ratio calculation unit 6, the feature amount distribution output from the feature amount distribution generation unit 4 is expressed by mixing the driving topic group stored in the driving topic database 5. The ratio (content ratio of each driving topic) is calculated (see FIG. 4). Note that the topic ratio can be obtained by executing only the E step among the processes of the E step and the M step that realize the estimation of the multimodal LDA.

<Probability calculation unit for allocation>
In the allocation probability calculation unit 7, as shown in FIG. 6A, the topic ratio obtained by the topic ratio calculation unit 6 and the image base distribution that is a part of the driving topic stored in the driving topic database 5 Based on the above, an allocation probability distribution is obtained.

Specifically, TP _k is the driving topic identified by the identifier k, w _i is the image feature amount belonging to the bin of the image feature amount distribution / image base distribution identified by the identifier i, and φ _ki is the driving topic TP _k. The region on the image having the image feature value w _i is the driving topic TP _k , where the i component of the multinomial distribution representing the image base distribution of _k , _k is the total number of topics, and θ _k is the ratio of the driving topic TP _k in the topic ratio. The probability P _ik derived from is calculated according to the equation (5).

Thus, for each index of the image feature amount distribution / image base distribution, the allocation probability distribution P _i = {P _i1 , which indicates which driving topic the local region having the image feature amount belonging to the index BIN _i is derived from. P _i2 ,... P _iK } is determined. However, the probability distribution in the figure shows a value before normalization (that is, the result of calculating only the numerator of the equation (5)) so that the sum of the probability distributions becomes 1.

<Emphasis area setting section>
As shown in FIG. 6B, the enhancement region setting unit 8 uses the feature amount calculated for each local region by the image feature calculation unit 42 to assign an allocation probability distribution (assignment probability) to which the feature amount belongs. According to the calculation result of the calculation unit 7, a driving topic is probabilistically assigned to the local region (region on the image) for which the feature amount is calculated. However, the assignment method may not be assigned probabilistically. For example, the driving topic having the maximum probability may be assigned according to the assignment probability.

  Further, among driving topics, those satisfying a predetermined presentation condition are regarded as salient topics, and an area on the image to which the salient topic is assigned is emphasized in a form (color, hatching pattern, etc.) previously associated with the salient topic. Generate highlighted data for display. Here, the presentation condition is that the topic is designated as a salient topic in advance. Further, it is assumed that the highlighting mode is defined in advance for each remarkable topic.

  Then, the highlight region setting unit 8 outputs the generated highlight display data to the display device 12 that performs image display based on the image data supplied from the in-vehicle camera 11. The display device 12 highlights an image based on the image data according to the highlight display data.

Here, a method of extracting a driving topic designated as a salient topic will be described.
First, characteristics unique to each driving topic are extracted using the driving basis distribution (see FIG. 5) constituting each driving topic. The likelihood ratio is used for the extraction. More specifically, w _j is a driving feature belonging to the bin identified by j, TP _k is a driving topic identified by the identifier k, φ _kj is a multinomial distribution j representing the driving base distribution of the driving topic TP _k. As a component, a likelihood ratio L _jk is obtained according to equation (6). However, the denominator in the equation (6) means that the sum of components other than k is taken for the component φ _kj of the multinomial distribution. The likelihood ratio L _jk is an index called “inverse document frequency (IDF)” indicating a degree that a certain feature is not included in another driving topic but is included in a specific driving topic.

FIG. 7 is a feature enhancement distribution in which only the likelihood ratio L _jk is greater than or _equal to a certain value (here, 5) in each driving topic. Here, in order to make the drawing easier to see, the accelerator, brake, steering, and speed are shown. Only shows.

  Based on this feature enhancement distribution, a driving topic that satisfies a predetermined condition is designated as a remarkable topic. Specifically, as the predetermined condition, for example, (1) in the accelerator opening histogram, the likelihood ratio of the bin where the value of the accelerator opening is 50% or more is 5 or more, and (2) in the speed histogram The likelihood ratio of the bin having a speed value of 80 km or more can be 5 or more. In the case of (1), the topic TP3 is designated as a salient topic in FIG. This remarkable topic can be interpreted as a driving topic (high acceleration topic) that appears when the accelerator is depressed. In the case of (2), the topic TP5 is designated as a notable topic in FIG. This salient topic can be interpreted as a driving topic (high speed traveling topic) that appears during high speed traveling or the like. That is, various driving contexts are expressed by individual salient topics or combinations of salient topics.

  Notable topics are not limited to this, for example, driving topics that appear when the brakes are depressed (high braking topics), driving topics that appear when driving at low speeds (low speed driving topics), and driving topics that appear when changing lanes (lane changes) Topic), a driving topic that can derive a meaningful interpretation from the feature enhancement distribution, such as a driving topic (interrupted topic) that appears when the vehicle has interrupted forward, may be used.

  FIG. 8 shows an example of an image displayed on the display device 12. FIG. 8A shows a case where “high brake topic” is designated as a prominent topic, and the driver steps on the brake so as not to collide with the preceding vehicle. (B) illustrates the situation when driving on a highway when “high speed driving topic” is designated as the salient topic. From these figures, it can be seen that in each driving context (significant topic), the area where the driver should be alerted is highlighted. This is because a specific driving feature amount and an image feature amount related to the specific driving feature amount are repeatedly generated and extracted as a remarkable topic.

  In FIG. 8, the case where there is one salient topic is illustrated for easy viewing of the drawing. However, when a plurality of salient topics are set, the areas corresponding to the salient topics are different from each other (for example, Color, hatching pattern).

<Effect>
As described above, in the driving support device 1, the time series of the driving behavior data is divided into partial series each representing some driving scene by the DDA which is an unsupervised division method, and the driving behavior is divided for each partial series. The driving context (driving topic) grasped from the feature amount distribution generated based on the data and the image data is extracted, and the local region associated with the extracted driving context is highlighted.

  As described above, according to the driving support device 1, instead of determining which of the driving scenes defined in advance is similar to the current situation, what kind of driving topic the driving scene is composed of Because the driving context is extracted by analyzing whether it is being used, it is possible to provide information according to the driving context instead of the predefined information for every driving scene that appears from the start to the end of driving Can do.

  In addition, by using DAA to divide the driving scene, the driving scene is extracted in a unit suitable for the driver's intuition. Therefore, the extracted driving scene is a vehicle situation including a stable driver's intention, that is, Thus, the specific driving context is strongly reflected, and the driving context can be extracted with high accuracy.

<Other embodiments>
As mentioned above, although embodiment of this invention was described, it cannot be overemphasized that this invention can take a various form, without being limited to the said embodiment.

  (1) In the above embodiment, driving behavior data and image data are used as the generation target of the feature amount distribution, but the present invention is not limited to this. For example, in addition to the position data related to the position of the host vehicle acquired by the Global Positioning System (GPS) and the inertial measurement unit (IMU), the data representing the time zone, the data representing the weather, etc. Any data may be used as the vehicle behavior data or the environmental data as long as it may affect the driving operation of the driver or the behavior of the vehicle.

  (2) In the above embodiment, image data is used as data used for information presentation. However, the present invention is not limited to this. For example, data representing the distance and direction to an object detected by various radars, or a known distance Distance image data detected by an image sensor or the like may be used. Even when these data are used, processing can be performed in the same manner as when normal image data is used by obtaining a local feature amount corresponding to the sensor to be used. As an example, in FIG. 9, (a) is an information presentation image when distance image data is used, and (b) shows that the preceding vehicle is the target of highlighting in a situation of traveling on a highway. ) Shows an information presentation image when normal image data is used for comparison. When distance image data is used, a local feature descriptor for a three-dimensional point group such as a SPIN feature amount or a SHOT feature amount can be used as a local feature.

  (3) In the above-described embodiment, “presented as a prominent topic” is used as the presentation condition. However, the present invention is not limited to this. For example, you may use the presentation conditions that a remarkable topic switches according to some conditions. In addition, the saliency topic may be designated so that the user can arbitrarily set it.

  (4) In the above embodiment, the information is visually presented by highlighting the local region to which the salient topic is assigned on the screen. However, the present invention is not limited to this. For example, an alarm sound is generated by a sound source that has been localized in the direction in which the local area to which the remarkable topic is assigned, that is, the direction in which the driver wants to be alerted, is presented auditorily, information is presented, Information may be presented tactilely, such as by vibrating a handle.

  (5) Each component of the present invention is conceptual and is not limited to the above embodiment. For example, the functions of one component may be distributed to a plurality of components, or the functions of a plurality of components may be integrated into one component. Further, at least a part of the configuration of the above embodiment may be replaced with a known configuration having the same function. In addition, at least a part of the configuration of the above embodiment may be added to or replaced with the configuration of the other embodiment.

  DESCRIPTION OF SYMBOLS 1 ... Driving assistance device 2 ... Driving behavior data collection part 3 ... Driving scene discretization part 4 ... Feature-value distribution generation part 5 ... Driving topic database 6 ... Topic ratio calculation part 7 ... Probability calculation part 8 for allocation 8 ... Emphasis area setting part DESCRIPTION OF SYMBOLS 9 ... Environmental data collection part 11 ... Car-mounted camera 12 ... Display apparatus 41 ... Driving | running feature calculation part 42 ... Image feature calculation part

Claims (7)

Driving behavior data collecting means (2) for repeatedly collecting driving behavior data relating to the driving behavior of the vehicle;
Environmental data collection means (9) for repeatedly collecting environmental data representing the environment around the vehicle;
Scene dividing means (3) for dividing the series of the driving behavior data collected by the driving behavior data collecting means into a plurality of partial series each representing some driving scene;
One or more types of feature quantities obtained from the driving behavior data collected by the driving behavior data collection means are driving feature quantities, and local feature quantities obtained from the environmental data collected by the environmental data collection means are environmental features. As a quantity, a feature quantity distribution generating means (4) for generating a feature quantity distribution representing the appearance frequency of the feature quantity for each partial series and for each feature quantity;
Using a plurality of feature quantity distributions generated by the feature quantity distribution generating means for each partial series as a feature quantity distribution group, using the co-occurrence between the driving feature quantity distribution and the environmental feature quantity distribution in the feature quantity distribution group And notification means (5, 6, 7, 8) for controlling notification related to the local area of the environmental data
A driving support apparatus comprising: The notification means includes
Topic storage means (5) for storing a driving topic group consisting of a plurality of types of driving topics used as a base distribution when expressing the feature quantity distribution group;
A ratio calculation means (6) for calculating a topic ratio, which is a mixture ratio when the feature quantity distribution group is expressed using the driving topic group, for each partial series;
For each partial series, based on the topic ratio and the driving topic group, for each of the environmental feature values, for assignment, the local region having the environmental feature value is stochastically derived from which driving topic. Probability distribution generation means (7) for generating a probability distribution;
Notification control means (8) for allocating the driving topic for each local region according to the environmental feature and the probability distribution and controlling notification regarding the local region to which the driving topic satisfying a preset presentation condition is allocated. When,
The driving support device according to claim 1, further comprising:   3. The driving support device according to claim 1, wherein when there are a plurality of driving topics that satisfy the presentation condition, the notification control unit performs the notification in a different manner for each driving topic.   The driving support apparatus according to any one of claims 1 to 3, wherein the notification control unit highlights the local region to be notified on an image based on the environmental data.   The driving assistance apparatus according to any one of claims 1 to 4, wherein the environmental data collection unit collects image data or distance image data as the environmental data.   6. The scene dividing means comprises a double segment analyzer (DAA: Double Articulation Analyzer) that performs segmentation by an unsupervised driving scene segmentation method using a double segment structure. The driving support device according to any one of the above.   The operation according to any one of claims 1 to 6, wherein the ratio calculation means obtains the topic ratio by a latent dirichlet allocation method (LDA) or a multimodal LDA. Support device.