JP3780655B2 - Driving behavior pattern recognition device - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to an apparatus for recognizing a movement behavior pattern of a driver who drives a vehicle.
[0002]
[Prior art]
As a technique for recognizing a driving behavior pattern from time series data such as an operation amount of a driver who drives a vehicle, a vehicle state quantity, etc., for example, “Nissan Cambridge Basic Research Annual 1996, Nissan Cambridge Basic Research, Nissan Research Research, Inc. There is a method described in. This prior art is an observation symbol for time-series data of categories, vehicle speed, vehicle longitudinal acceleration, steering angle, and steering angular velocity that recognize driving behavior patterns such as right turn, left turn, braking stop, lane change, etc. Recognize each series using the maximum likelihood method based on the hidden Markov model, with a very high positive recognition rate (the ratio of the number of sample data in which the correct category appears as the recognition result in the total number of sample data) ) Has been reported.
[0003]
[Problems to be solved by the invention]
However, such a conventional method has a problem that the recognition accuracy may be lowered while an extremely high positive recognition rate is obtained.
Here, the positive recognition rate Corr and the recognition accuracy Acc are indexes for evaluating the performance of the model calculated by the following equations, respectively.
Corr = H / N × 100 (%)
Acc = (HI) / N x 100 (%)
N: Number of sample data H: Number of data in which a correct category appears as a recognition result I: Number of times an error has occurred in which another category is erroneously inserted
That is, in the conventional method, the driving behavior pattern to be recognized is a series of actions (hereinafter referred to as “maneuver”) performed under the driver's intention such as right turn and left turn, and is hidden at the maneuver level. A Markov model is built and parameters are estimated and recognized. In such a method, for example, in the case of a right turn or a left turn, both vehicles are used at the time of deceleration when entering an intersection, which is the initial stage of each maneuver. Similarity is high in that the data is decelerated. When such data is given to the model as recognition data, an insertion error is likely to occur, and the recognition accuracy is lowered.
[0005]
The present invention has been made paying attention to such conventional problems, and an object of the present invention is to provide a technique capable of recognizing a maneuver with high recognition accuracy.
[0006]
[Means for Solving the Problems]
In order to solve the above-described problems, the present invention regards a driving behavior pattern as a hierarchical structure of sub-patterns that are performed to perform a driving behavior pattern that is a superordinate concept and a driving behavior pattern of a lower structure, and a driving behavior pattern. Is a subdivision of each driving behavior pattern to be recognized based on the idea that it can be expressed as a discrete state transition model of sub-patterns, and different driving behaviors Parameter estimation that estimates the statistical features necessary to represent subpatterns that are treated as the same if they have high similarity as hidden Markov models using the driving behavior data of each driving behavior pattern as an observation symbol sequence And the probability that the driving behavior data is output from each driving behavior sub-pattern model is estimated by the parameter estimating means. Driving behavior sub-pattern recognition means for generating and outputting a sub-pattern sequence with a sub-pattern having the highest output probability as a recognition result, and a sub-pattern of each driving behavior pattern. Matching the driving behavior pattern template describing the temporal context, the sub-pattern permutation in the sub-pattern sequence output from the driving behavior sub-pattern recognition means, and the sub-pattern permutation of each driving behavior pattern described in the driving behavior pattern template And a driving action pattern recognition means for outputting a driving action pattern that matches the both as a recognition result is provided.
[0007]
Hereinafter, the operation of the present invention will be described.
According to the first aspect of the present invention, the parameter estimation and recognition of the driving action pattern are performed in units of sub-patterns, which are divided into the driving action patterns and those having high similarity among different driving action patterns, and the recognition result When the driving pattern is recognized by matching the sub-pattern sequence with the sub-pattern template, when part of the time-series data overlaps in the feature space even though they are different driving behavior patterns In addition, erroneous driving behavior patterns are not output as recognition results, and driving behavior patterns can be recognized with high recognition accuracy.
[0008]
Further, in the invention according to claim 2, when there is no sub-pattern sequence that matches the recognition result of the sub-pattern sequence and the sub-pattern template, there is no corresponding driving behavior pattern as the recognition result. By outputting, it is possible to prevent an erroneous driving behavior pattern from being output and to recognize the driving behavior pattern with high recognition accuracy.
[0009]
DETAILED DESCRIPTION OF THE INVENTION
DESCRIPTION OF EMBODIMENTS Hereinafter, an embodiment of a driving behavior pattern recognition device according to the present invention will be described in detail with reference to the accompanying drawings.
FIG. 1 is a block diagram showing a schematic configuration of an embodiment of a driving behavior pattern recognition device according to the present invention, and FIG. 2 is a block diagram showing a functional configuration of an embodiment of a driving behavior pattern recognition device according to the present invention. is there.
[0010]
First, the configuration will be described with reference to FIG.
Driving action data input means 1 for inputting driving action data that is time series data including at least one of driving operation amount and vehicle state quantity, and each driving action pattern to be recognized is subdivided in time series. In addition, statistical features necessary to express sub-patterns that are treated as the same among different driving behavior patterns that are the same as hidden Markov models, and driving behavior data of each driving behavior pattern as observation symbol series The parameter estimation means 3 estimated using the driving behavior data, and the probability that each driving behavior sub-pattern is output from the driving behavior data is calculated using the parameter estimated by the parameter estimation means 3 and the output probability calculation algorithm. Driving behavior sub-pattern recognition that generates and outputs a sub-pattern sequence with a high sub-pattern as a recognition result Step 7, a driving behavior pattern template 5 that describes the temporal order of sub-patterns in each driving behavior pattern, a sub-pattern permutation in the sub-pattern series output from the driving behavior sub-pattern recognition means 7, and a driving behavior pattern template The driving behavior pattern recognition means 9 is provided to check the sub-pattern permutation of each driving behavior pattern described in 5 and output the matching driving behavior pattern as a recognition result.
[0011]
Hereinafter, as an example for explaining an embodiment of the present invention, a recognition system that recognizes five types of driving behavior patterns of right turn, left turn, braking stop, lane change, and overtaking will be described. Of these, there are two types of lane change: lane change on the right side of the vehicle and lane change on the left side of the vehicle. Therefore, consider the recognition of six types of driving behavior patterns.
[0012]
The driving behavior data is data relating to driving behavior such as an accelerator pedal opening degree and a steering angle as a driver's operation amount, and a vehicle state amount such as a host vehicle speed and a host vehicle longitudinal acceleration. And is stored in the driving action data storage memory 15 mounted on the computer 13. Hidden Markov model parameter estimation and driving behavior pattern recognition are performed by using time series data stored in the driving behavior data storage memory 15 as the observed symbol series O.
[0013]
The category to be recognized is modeled by a hidden Markov model λ = {π, A, B}. Here, π represents an initial occurrence probability set of each state, A represents an inter-state transition probability set, and B represents an output probability set. When the hidden Markov model is applied as a recognition system, it is first necessary to obtain πi, Ai, Bi, which are statistical feature quantities, for a category to be recognized.
[0014]
FIG. 3 shows an example in which driving behavior patterns to be recognized are classified into sub-patterns. In the present invention, driving behavior patterns are considered as discrete state transition expressions of driving behavior sub-patterns in this way. . Note that “do-nothing” in FIG. 3 means a sub-pattern model indicating a steady driving state in which there is no change in driving operation such as an accelerator pedal, steering, etc., and this state has a time gap between the sub-patterns. Although it is set in order to be recognized when it exists, it is not necessarily present in all driving behavior data, so it is in a transition format that allows jumping. Here, even if different driving action patterns are used, those having high similarity such as braking, right turning, left turning, and acceleration are set as a common driving action sub-pattern.
[0015]
The pre-processing unit 17 extracts the driving behavior sub-pattern of the category to be learned from the driving behavior data stored in the driving behavior data storage memory 15 so that the parameter of the hidden Markov model can be estimated, and stores the driving behavior sub-pattern data. It is temporarily stored in the memory 19. FIG. 4 shows an example of right turn as an example of dividing driving action data into driving action sub-patterns.
[0016]
In the parameter estimation unit 21, the initial probability π, the inter-state transition probability A, and the output probability B, which are the parameters of the hidden Markov model, are set as driving behavior sub-pattern data for each driving behavior sub-pattern that is recognized using the hidden Markov model. Using the observation symbol sequence in units of driving behavior sub-patterns stored in the storage memory 19, it is estimated that the probability P (λ | O) output from the driving behavior data is maximized.
[0017]
A Baum-Welch algorithm is generally used for parameter estimation of a hidden Markov model. The Baum-Welch algorithm, the Forward algorithm and the Viterbi algorithm described later are described in, for example, the document “X. D. Huang, Y. et al. Ariki, and M.M. A. Jack. Hidden Markov Models for Speech Recognition. The contents are generally well known by Edinburgh University Press, Edinburgh, 1990 ”, and a specific calculation formula and procedure are omitted here. When the learning by the Baum-Welch algorithm converges, the estimated parameters are stored in the parameter storage memory 23.
[0018]
The driving action sub-pattern recognition execution unit 25 uses the Forward algorithm or Viterbi algorithm, which is a general hidden Markov model output probability calculation algorithm, to calculate the time series of driving action patterns held in the driving action data storage memory 15. The output probability P (O | λ) of each sub-pattern is calculated using the parameters held in the data and parameter storage memory 23, and the sub-pattern with the highest output probability is output as the recognition result.
[0019]
The sub pattern series generation unit 27 generates a sub pattern series based on the recognition result of the driving action sub pattern recognition execution unit 25. At this time, “do-nothing” is excluded. FIG. 5 shows an example of generating a sub-pattern sequence when turning right.
[0020]
In the sub-pattern template 29, a sub-pattern series in each driving action pattern is described. FIG. 6 shows a description content example of the sub-pattern template.
[0021]
The collation unit 31 collates the sub pattern series generated by the sub pattern series generation unit 27 with the sub pattern series of each driving action pattern described in the sub pattern template 29, and there is a matching driving action pattern. In this case, the matching driving behavior pattern is stored in the recognition result output memory 33, and if there is no matching driving behavior pattern, the absence is stored.
[0022]
【The invention's effect】
As described above in detail, according to the present invention, in recognition of a driving action pattern using a hidden Markov model, parameter estimation and recognition are performed in units of driving action sub-patterns constituting the driving action pattern, and driving is performed. A driving behavior pattern template that describes the temporal context between the driving behavior sub-patterns that make up the behavior pattern is provided, and the driving behavior pattern is recognized by matching the sub-pattern sequence with the sub-pattern template as a recognition result. As a result, misrecognition due to insertion errors is reduced, and driving behavior patterns can be recognized at a driving behavior pattern level with high recognition accuracy.
[Brief description of the drawings]
FIG. 1 is a block diagram showing a schematic configuration of an embodiment of a driving behavior pattern recognition apparatus according to the present invention.
FIG. 2 is a block diagram showing a functional configuration of an embodiment of a driving behavior pattern recognition apparatus according to the present invention.
FIG. 3 is a diagram illustrating a relationship example between a driving behavior pattern to be recognized and a driving behavior sub-pattern.
FIG. 4 is a diagram illustrating an example of time-series data of driving behavior patterns and an example of division into driving behavior sub-patterns (when turning right).
FIG. 5 is a diagram illustrating an example of generating a sub-pattern series from a recognition result (when turning right).
FIG. 6 is a diagram illustrating a description content example of a sub-pattern template.
[Explanation of symbols]
DESCRIPTION OF SYMBOLS 1 Driving action data input means 3 Parameter estimation means 5 Driving action pattern template 7 Driving action sub-pattern recognition means 9 Driving action pattern recognition means 11 Various sensors 13 Computer 15 Driving action data storage memory 17 Preprocessing part 19 Driving action sub-pattern data storage Memory 21 Parameter estimation unit 23 Parameter storage memory 25 Driving action sub-pattern recognition execution unit 27 Sub-pattern series generation unit 29 Sub-pattern template 31 Collation unit 33 Recognition result output memory

Claims (2)

Driving action data input means for inputting driving action data that is time-series data including at least one of a driving operation amount and a vehicle state quantity;
Statistics required for expressing sub-patterns that are treated as the same for sub-patterns that are classified as time-series and that have similarities even in different driving behavior patterns as hidden Markov models Parameter estimation means for estimating a characteristic feature amount using driving behavior data of each driving behavior pattern as an observation symbol sequence,
The probability that the driving behavior data is output from each driving behavior sub-pattern represented by the parameter estimated by the parameter estimation means is calculated using an output probability calculation algorithm, and the sub-pattern with the highest output probability is recognized. As a result, driving action sub-pattern recognition means for generating and outputting a sub-pattern sequence,
A driving behavior pattern template that describes the temporal context of sub-patterns in each driving behavior pattern;
A sub-pattern permutation in the sub-pattern sequence output from the driving behavior sub-pattern recognition means is compared with a sub-pattern permutation of each driving behavior pattern described in the driving behavior pattern template, and a driving behavior pattern in which both match is determined. A driving behavior pattern recognition device comprising driving behavior pattern recognition means for outputting a recognition result. The driving behavior pattern recognition device according to claim 1,
The driving behavior pattern recognizing means, when the matching pattern does not exist in the matching of the sub-pattern sequence output as a recognition result from the driving behavior sub-pattern recognition means and the driving behavior sub-pattern template, the corresponding driving A driving behavior pattern recognition apparatus that outputs a recognition result indicating that there is no behavior pattern.

Images