JPH1199846A - Operating action pattern recognizing device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To recognize an operating action pattern with high accuracy by estimating and recognizing a parameter by units of operating action sub-patterns, and recognizing an operation action pattern by collation between the sub-pattern series as a recognition result and a sub-pattern template. SOLUTION: This device comprises an operating action data input means 1 for inputting operating action data, and the statistical characteristic amount is estimated from the operating action data of each operating action pattern by a parameter estimating means 3. Further the device comprises an operating action sub-pattern recognizing means 7, wherein the probability of outputting each operating action sub-pattern is calculated from the operating action data, a sub-pattern having the highest probability of output is taken as a recognition result, and a sub-pattern series is generated. By an operating action pattern recognizing means 9, a sub-pattern permutation in the sub-pattern series is collated with a sub-pattern permutation of each operating action pattern read out from an operating action pattern template 5 to output an operating action pattern as a recognition result.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an apparatus for recognizing a motor behavior pattern of a driver driving a vehicle.

[0002]

2. Description of the Related Art As a technique for recognizing a driving behavior pattern from time-series data such as an operation amount of a driver driving a vehicle and a vehicle state quantity, for example, "Nissan Cambr"
idge Basic Research Annua
l Report 1996, Nissan Camb
Ridge Basic Research, Niss
an Research and Development
ent Inc. There is a method described in. This conventional technique uses time-series data of a category, a subject vehicle speed, a subject vehicle longitudinal acceleration, a steering angle, and a steering angular speed to recognize a driving behavior pattern such as a right turn, a left turn, a braking stop, and a lane change when driving a car as an observation symbol. This is a method of performing recognition using the maximum likelihood method based on a hidden Markov model for each series. An extremely high positive recognition rate (the ratio of the number of sample data in which the correct category appears as a recognition result to the total sample data) ) Has been reported.

[0003]

However, in such a conventional method, there is a problem that while a very high correct recognition rate is obtained, the recognition accuracy may be lowered. Here, the correct recognition rate Corr and the recognition accuracy Acc are indices for evaluating the performance of the model calculated by the following equations, respectively. Corr = H / N × 100 (%) Acc = (HI) / N × 100 (%) N: Number of sample data H: Number of data in which correct category appears as recognition result I: Another category is erroneously inserted Number of times an error occurred

That is, in the above-mentioned conventional method, a driving action pattern to be recognized as a target is a series of actions (hereinafter abbreviated as “maneuver”) performed under the driver's intention such as right turn and left turn, and is referred to as maneuver. Although a hidden Markov model is constructed at the level to estimate and recognize parameters, for example, in the case of a right turn or left turn, for example, when decelerating at the intersection approach, which is the initial stage of each maneuver, Both have a high similarity in that the vehicle is decelerated. When such data is given to the model as recognition data, insertion errors are likely to occur, and the recognition accuracy is reduced. Was.

The present invention has been made in view of such conventional problems, and has as its object to provide a technique capable of recognizing a maneuver with high recognition accuracy.

[0006]

In order to solve the above-mentioned problems, the present invention relates to a sub-pattern performed for performing a driving behavior pattern of a superordinate concept and a driving behavior pattern of a lower structure. Based on the idea that driving behavior patterns can be expressed as discrete state transition models of subpatterns, each driving behavior pattern to be recognized was subdivided in time series. Sub-patterns that have the same similarity in different driving behavior patterns
Parameter estimation means for estimating a statistical feature amount required to be represented as a hidden Markov model using driving action data of each driving action pattern as an observed symbol sequence;
Driving behavior data is calculated from each driving behavior sub-pattern model using a parameter estimated by the parameter estimating means and an output probability calculation algorithm, and the sub-pattern having the highest output probability is used as a recognition result,
Driving behavior sub-pattern recognition means for generating and outputting a sub-pattern sequence, a driving behavior pattern template describing a temporal context of the sub-pattern in each driving behavior pattern, and a sub-pattern sequence output from the driving behavior sub-pattern recognition means A driving behavior pattern recognizing means for comparing the sub-pattern permutation in the sub-pattern permutation of each driving behavior pattern described in the driving behavior pattern template and outputting a driving behavior pattern that matches both as a recognition result is provided. .

Hereinafter, the operation of the present invention will be described. Claim 1
In the invention described in the above, the parameter estimation and recognition of the driving behavior pattern are performed by subdividing the driving behavior pattern, and those having high similarity between different driving behavior patterns are performed in units of common sub-patterns, and the sub-results of the recognition result Recognition of the driving behavior pattern by matching the pattern sequence with the sub-pattern template enables false recognition even when part of the time-series data overlaps in the feature space despite different driving behavior patterns. The driving behavior pattern thus output is not output as a recognition result, and the driving behavior pattern can be recognized with high recognition accuracy.

According to the second aspect of the present invention, when there is no matching sub-pattern sequence when the sub-pattern sequence is matched with the sub-pattern template as the recognition result, the corresponding driving behavior pattern exists as the recognition result. By outputting that no driving action pattern is output, an erroneous driving action pattern is prevented from being output, and the driving action pattern can be recognized with high recognition accuracy.

[0009]

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a block diagram showing an embodiment of a driving behavior pattern recognition apparatus according to the present invention. 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.
1 is a block diagram showing a functional configuration of a driving behavior pattern recognition device according to an embodiment of the present invention.

First, the configuration will be described with reference to FIG. Driving action data input means 1 for inputting driving action data as time series data including at least one of a driving operation amount and a vehicle state quantity, and each driving action pattern to be recognized is subdivided in time series. Even if the driving behavior patterns are different, the ones with high similarity are treated as the same.The statistical features required to represent the sub-pattern as a hidden Markov model are represented by the driving behavior data of each driving behavior pattern in the observed symbol sequence. The parameter estimating means 3 for estimating the output probabilities and the probabilities that each driving action sub-pattern is output from the driving action data are calculated using the parameters estimated by the parameter estimating means 3 and the output probability calculating algorithm, and Driving behavior sub-pattern recognition that generates and outputs a sub-pattern sequence using high sub-patterns as recognition results A stage 7, a driving behavior pattern template 5 describes the temporal context of the sub-patterns in each driving behavior patterns, driving behavior sub-pattern recognition means 7
The sub-pattern permutation in the sub-pattern sequence output from the sub-pattern sequence is compared with the sub-pattern permutation of each driving behavior pattern described in the driving behavior pattern template 5, and the driving behavior pattern in which both match is output as a recognition result. The configuration is such that pattern recognition means 9 is provided.

Hereinafter, as an example for describing an embodiment of the present invention, a recognition system for recognizing five types of driving action patterns of right turn, left turn, braking stop, lane change, and overtaking will be described. Of these, two types of lane change, lane change to the right lane of the own vehicle and lane change to the left lane of the own vehicle, are considered, so that they are treated as different driving behavior patterns. Therefore, consider the recognition of six types of driving behavior patterns.

The driving behavior data includes, for example, an accelerator pedal opening degree and a steering angle as a driver's operation amount.
The vehicle state quantity is, for example, data relating to driving behavior such as the vehicle speed and the vehicle longitudinal acceleration. These are measured by various sensors 11 and stored in the driving behavior data storage memory 15 mounted on the computer 13. The estimation of the parameters of the hidden Markov model and the recognition of the driving behavior pattern are performed by using the time-series data stored in the driving behavior data storage memory 15 as the observation symbol sequence O.

The category to be recognized is represented by a hidden Markov model λ = {π, A, B}. Here, π indicates an initial occurrence probability set of each state, A indicates an inter-state transition probability set, and B indicates an output probability set. When applying the Hidden Markov Model as a recognition system, it is necessary to first obtain the statistical features πi, Ai, and Bi of the category to be recognized.

FIG. 3 shows an example in which the driving behavior patterns to be recognized are classified into sub-patterns. In the present invention, the driving behavior patterns are divided into discrete state transitions of the driving behavior sub-patterns. Think as an expression. In addition,
“Do-noting” 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 and steering, and this state has a temporal gap between sub-patterns. Although it is set to be recognized in the case, it is not always present in all the driving behavior data, so that the transition form is such that jumping is possible. Here, even with different driving behavior patterns,
Those having a high similarity, such as braking, turning right, turning left, and accelerating, have a common driving behavior sub-pattern.

The preprocessing unit 17 stores driving behavior sub-patterns of the category to be learned in the driving behavior data storage memory 1 so that the parameters of the hidden Markov model can be estimated.
5 and temporarily stored in the driving behavior sub-pattern data storage memory 19. FIG. 4 shows an example at the time of a right turn as an example of dividing the driving behavior data into driving behavior sub-patterns.

The parameter estimating unit 21 calculates the initial probability π, the transition probability between states A, and the output probability B, which are the parameters of the hidden Markov model, for each of the driving behavior subpatterns to be recognized using the hidden Markov model. Using the observed symbol sequence of the driving behavior sub-pattern unit stored in the sub-pattern data storage memory 19, estimation is performed so that the probability P (λ | O) output from the driving behavior data is maximized.

For estimating the parameters of the hidden Markov model, a Baum-Welch algorithm is generally used. The Baum-Welch algorithm,
And Forward algorithm, Vite described later
The rbi algorithm is described in, for example, the document “X. D. Hua
ng, Y. Arik, and M.S. A. Jack. H
iden Markov Models for S
peech Recognition. Edinbu
rgh University Press, Edin
Burgh, 1990], the contents of which are well known in general, and specific calculation formulas and procedures are omitted here. When learning by the Baum-Welch algorithm converges, the estimated parameters are stored in the parameter storage memory 23.

The driving behavior sub-pattern recognition execution unit 25 executes the Forward algorithm, which is an algorithm for calculating the output probability of a general hidden Markov model, or Vi.
Using the terbi algorithm, the output probability P (O | λ) of each sub-pattern using the time-series data of the driving behavior pattern stored in the driving behavior data storage memory 15 and the parameters stored in the parameter storage memory 23.
Is calculated, and the sub-pattern having the highest output probability is output as a recognition result.

The sub-pattern sequence generation unit 27 generates a sub-pattern sequence based on the recognition result of the driving behavior sub-pattern recognition execution unit 25. At this time, "do-nothin
g "is excluded. FIG. 5 shows an example of generating a sub-pattern sequence when turning right.

The sub-pattern template 29 describes a sub-pattern sequence in each driving behavior pattern. FIG. 6 shows an example of the description contents of the sub-pattern template.

The collating unit 31 collates the sub-pattern sequence generated by the sub-pattern sequence generating unit 27 with the sub-pattern sequence of each driving behavior pattern described in the sub-pattern template 29, and matches the matching driving behavior pattern. Is stored in the recognition result output memory 33, and the matched driving behavior pattern is stored in the recognition result output memory 33 when there is no matching driving behavior pattern.

[0022]

As described above in detail, according to the present invention, in recognition of a driving behavior pattern using a hidden Markov model, estimation and recognition of a parameter are performed in units of driving behavior sub-patterns constituting the driving behavior pattern. In addition, a driving behavior pattern template that describes the temporal relationship between driving behavior sub-patterns that constitute the driving behavior pattern is provided, and the recognition of the driving behavior pattern is performed by comparing the sub-pattern template with the sub-pattern sequence that is the recognition result. Since the configuration is performed, erroneous recognition due to an insertion error is reduced, and it is possible to recognize a driving behavior pattern 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 device according to the present invention.

FIG. 2 is a block diagram showing a functional configuration of a driving behavior pattern recognition device according to an embodiment of 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 showing an example of time-series data of a driving action pattern and an example of division into driving action sub-patterns (at the time of turning right).

FIG. 5 is a diagram showing an example of generating a sub-pattern sequence from a recognition result (when turning right).

FIG. 6 is a diagram showing an example of description contents of a sub-pattern template.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Driving behavior data input means 3 Parameter estimation means 5 Driving behavior pattern template 7 Driving behavior sub-pattern recognition means 9 Driving behavior pattern recognition means 11 Various sensors 13 Computer 15 Driving behavior data storage memory 17 Preprocessing unit 19 Driving behavior sub-pattern data storage Memory 21 Parameter estimation unit 23 Parameter storage memory 25 Driving behavior sub-pattern recognition execution unit 27 Sub-pattern sequence generation unit 29 Sub-pattern template 31 Collation unit 33 Recognition result output memory

Claims (2)

[Claims] 1. A driving action data input means for inputting driving action data which is time series data including at least one of a driving operation amount and a vehicle state quantity, and subdivides each driving action pattern to be recognized in time series. The statistical features required to represent a sub-pattern as a hidden Markov model, which is treated as the same even if the driving behavior patterns are different and have a high similarity, are used as the driving behavior data of each driving behavior pattern. Parameter estimation means for estimating using the observed symbol sequence, the driving behavior data, the probability of being output from each driving behavior sub-pattern represented by the parameter estimated by the parameter estimation means, using an output probability calculation algorithm The sub-pattern with the highest output probability is used as the recognition result, and a sub-pattern sequence is generated and output. Driving behavior sub-pattern recognition means, a driving behavior pattern template describing the temporal context of sub-patterns in each driving behavior pattern, a sub-pattern permutation in a sub-pattern sequence output from the driving behavior sub-pattern recognition means, and A driving behavior pattern recognizing unit that performs sub-pattern permutation of each driving behavior pattern described in the driving behavior pattern template and outputs a driving behavior pattern that matches both as a recognition result. Recognition device. 2. The driving behavior pattern recognition device according to claim 1, wherein the driving behavior pattern recognition means includes a sub-pattern sequence output as a recognition result from the driving behavior sub-pattern recognition means and the driving behavior sub-pattern template. A driving behavior pattern recognition device, which outputs, as a recognition result, that there is no corresponding driving behavior pattern when there is no matching pattern in the collation of (1).