Method, system and apparatus for recognizing continuous driving style

Patent metadata

Field	Value
Publication number	US-12377860-B2
Application number	US-202418429787-A
Country	US
Kind code	B2
Filing date	Feb 1, 2024
Priority date	Aug 30, 2023
Publication date	Aug 5, 2025
Grant date	Aug 5, 2025

Abstract

A method, system and apparatus for recognizing continuous driving style are provided, which involve the intelligent vehicles field. The method comprises: collecting multi-dimensional driving data from a plurality of drivers in daily driving scenarios, segmenting the driving data to obtain a plurality of driving segments, calculating statistical features of each driving segment to determine high-dimensional continuous driving statistical features, reducing dimensionality of the driving statistical features to generate common factors for each driving segment, representing each driving segment with a driving word based on the common factors, representing all the driving segments of a target driver as a driving word sequence, constructing a hierarchical latent model of driving behavior, and inputting the driving word sequence into the hierarchical latent model of driving behavior to determine the continuous driving style of the target driver.

First claim

What is claimed is:

1. A method for recognizing continuous driving style, comprising:
collecting multi-dimensional driving data from a plurality of drivers in daily driving scenarios, wherein the multi-dimensional driving data comprises vehicle state data and driver operation data, wherein the vehicle state data comprises speed, longitudinal acceleration, lateral acceleration and yaw rate, and the driver operation data comprises throttle pedal position, brake pressure, and steering wheel angle;
segmenting the multi-dimensional driving data to obtain a plurality of driving segments;
calculating statistical features of the multi-dimensional driving data for each driving segment to determine high-dimensional continuous driving statistical features for all the driving segments;
reducing dimensionality of the high-dimensional continuous driving statistical features to generate common factors for each driving segment;
representing each driving segment with a driving word based on the common factors, and representing all the driving segments of a target driver as a driving word sequence based on driving words; wherein the driving words follow a first categorical distribution with basic driving styles as parameters, and the basic driving styles comprise an aggressive driving style and a moderate driving style; and
inputting the driving word sequence into a hierarchical latent model of driving behavior, and outputting, by the hierarchical latent model, the continuous driving style of the target driver; wherein the hierarchical latent model of the driving behavior is constructed based on the basic driving styles and the driving word corresponding to each driving segment, and the continuous driving style is mixture proportions of the basic driving styles learned from the driving word sequence.

2. The method of claim 1 , wherein the segmenting the multi-dimensional driving data to obtain a plurality of driving segments comprises:
segmenting time sequence of the multi-dimensional driving data by employing a time window, to determine the plurality of driving segments.

3. The method of claim 1 , wherein the reducing dimensionality of the high-dimensional continuous driving statistical features to generate common factors for each driving segment comprises:
reducing the dimensionality of the high-dimensional continuous driving statistical features by employing factor analysis, to generate the common factors for each driving segment, wherein a formula of the factor analysis is: Y=AF+e, wherein Y represents the driving statistical features, A denotes a factor loading matrix, F denotes a common factor matrix and e is an error term.

4. The method of claim 1 , wherein the representing each driving segment with a driving word based on the common factors comprises:
dividing common factors of all the driving segments, by using histogram statistics, into a plurality of intervals with an equal probability; and
constructing the driving words based on a number of the common factors and a number of the intervals.

5. The method of claim 1 , wherein a construction process of the hierarchical latent model of the driving behavior comprises:
utilizing a latent variable to represent the basic driving styles, and constructing a three-layer driving behavior model by using a probabilistic model based on the basic driving styles and the driving word corresponding to each driving segment,
wherein the three-layer driving behavior model comprises a driver layer, a basic driving style layer, and a driving word layer, and the three-layer driving behavior model is the hierarchical latent model of the driving behavior;
the basic driving styles follow a second categorical distribution with the continuous driving style of the target driver as a parameter;
the basic driving styles corresponding to each driving segment are sampled from a driver layer-basic driving style layer categorical distribution, to generate a basic driving style sequence corresponding to all the driving segments; and
based on the basic driving style sequence, the driving word corresponding to each driving segment is sampled from a basic driving style layer-driving word layer categorical distribution.

6. A system for recognizing continuous driving style, comprising:
a multi-dimensional driving data acquisition module, configured to collect multi-dimensional driving data from a plurality of drivers in daily driving scenarios, wherein the multi-dimensional driving data comprises vehicle state data and driver operation data, wherein the vehicle state data comprises speed, longitudinal acceleration, lateral acceleration and yaw rate, and the driver operation data comprises throttle pedal position, brake pressure, and steering wheel angle;
a segmentation module, configured to segment the multi-dimensional driving data to obtain a plurality of driving segments;
a driving statistical feature determination module, configured to calculate statistical features of the multi-dimensional driving data for each driving segment to determine high-dimensional continuous driving statistical features for all the driving segments;
a dimensionality reduction module, configured to reduce dimensionality of the high-dimensional continuous driving statistical features to generate common factors for each driving segment;
a driving word computation module, configured to represent each driving segment with a driving word based on the common factors, and represent all the driving segments of a target driver as a driving word sequence based on driving words; wherein the driving words follow a first categorical distribution with basic driving styles as parameters, and the basic driving styles comprise an aggressive driving style and a moderate driving style; and
a driving style recognition module, configured to input the driving word sequence into a hierarchical latent model of driving behavior, and output, by the hierarchical latent model, the continuous driving style of the target driver; wherein the hierarchical latent model of the driving behavior is constructed based on the basic driving styles and the driving word corresponding to each driving segment, and the continuous driving style is mixture proportions of the basic driving styles learned from the driving word sequence.

7. The system of claim 6 , wherein the driving word computation module comprises:
an equally dividing unit, configured to divide common factors of all the driving segments, by using histogram statistics, into a plurality of intervals with an equal probability; and
a driving word construction unit, configured to construct driving words based on a number of the common factors and a number of the intervals.

8. The system of claim 6 , wherein a construction process of the hierarchical latent model of the driving behavior comprises:
a driving behavior hierarchical latent model construction unit, configured to utilize a latent variable to represent the basic driving styles and construct a three-layer driving behavior model by using a probabilistic model based on the basic driving styles and the driving word corresponding to each driving segment;
wherein the three-layer driving behavior model comprises a driver layer, a basic driving style layer and a driving word layer, and the three-layer driving behavior model is the hierarchical latent model of the driving behavior;
the basic driving styles follow a second categorical distribution with the continuous driving style of the target driver as a parameter;
the basic driving styles corresponding to each driving segment are sampled from a driver layer-basic driving style layer categorical distribution to generate a basic driving style sequence corresponding to all the driving segments; and
based on the basic driving style seq

Assignees

• Beijing Institute Tech

Inventors

• Zhang Chaopeng
• Xi Junqiang
• Wang Wenshuo
• Wei Yao
• Chen Zhaokun
• Zhou Zikun

Classifications

• G06F17/18 Primary

  for evaluating statistical data {, e.g. average values, frequency distributions, probability functions, regression analysis (forecasting specially adapted for a specific administrative, business or logistic context G06Q10/04)} · CPC title

• B60W2540/30

  Driving style · CPC title

• G07C5/008

  communicating information to a remotely located station (transmission systems for measured values G08C) · CPC title

• B60W2556/05

  Big data · CPC title

• B60W2540/18

  Steering angle · CPC title