Method for evaluating driving risk level in tunnel based on vehicle bus data and system therefor

Patent metadata

Field	Value
Publication number	US-12570281-B2
Application number	US-202519223082-A
Country	US
Kind code	B2
Filing date	May 30, 2025
Priority date	Jun 5, 2024
Publication date	Mar 10, 2026
Grant date	Mar 10, 2026

Abstract

A method for evaluating a driving risk level in a tunnel based on vehicle bus data and a system therefor are provided. The method uses the Controller Area Network (CAN) bus data collected in a vehicle driving process, designs and extracts a driving risk characteristic feature index reflecting the driving behavior of a driver through a sliding time window method, writes a feature codebook to symbolize an extracted sequence feature, and then randomly samples all of the samples, and based on the sampled symbolic data, using a Latent Dirichlet Allocation (LDA) theme model to evaluate the driving risk level. The training method of the model is to acquire the optimal number of risk levels by evaluating the perplexity and the coherence scores, and to analyze the driving risk of the driving data in all of the samples.

First claim

The invention claimed is:

1 . A method for evaluating a driving risk level in a tunnel based on vehicle bus data, comprising:
step S1, acquiring bus data in a vehicle driving process in the tunnel, and extracting driving risk characteristic feature indexes by using a sliding window method according to the bus data to obtain a plurality of driving risk units; step S2, writing a feature index codebook according to the driving risk characteristic feature indexes, and obtaining a word sequence of each driving risk unit according to the feature index codebook; and step S3, using a Latent Dirichlet Allocation (LDA) model to analyze a risk type of each driving risk unit according to the word sequence to evaluate the driving risk level, wherein the driving risk level comprises normal driving, lateral risks, vertical risks; in response to the driving risk level of vertical risks, determining, by an advanced driving assistance system mounted in a current vehicle and in communication with the LDA model, that there is no situation of sudden braking of a preceding vehicle or abrupt lane intrusion by adjacent vehicles according to video data collected by a front camera of the current vehicle, outputting, by the advanced driving assistance system, a warning message to remind a driver of the current vehicle; in response to the driving risk level of lateral risks, determining, by the advanced driving assistance system, that there is no need for lanes change or steering maneuvers according to the video data collected by the front camera of the current vehicle, outputting, by the advanced driving assistance system, the warning message to remind the driver of the current vehicle; wherein in step S1, the bus data in the vehicle driving process comprises five driving behavior features of a steering wheel angle, a steering wheel angle change rate, an accelerator pedal position, a brake pedal position and a brake pedal change rate; each driving behavior feature has four driving risk characteristic feature indexes of an average value of the driving behavior feature within a time window, a standard deviation of the driving behavior feature within the time window, a maximum value of the driving behavior feature within the time window, and a minimum value of the driving behavior feature within the time window; and the step of writing the feature index codebook according to the driving risk characteristic feature indexes comprises: dividing values of the driving risk characteristic feature indexes into several ranges, so that each range corresponds to a word, which means encoding each driving risk characteristic feature index according to data distribution and domain knowledge to obtain the feature index codebook, wherein 20 driving risk characteristic feature indexes of each driving risk unit correspond to 20 words, respectively, and each driving risk unit is represented by a word sequence:

d
m

=

[

w

m
⁢
1

,

w

m
⁢
2

,
…
,

w
mn

]

wherein d m denotes a word sequence of an m-th risk unit, and W mn denotes an n-th word in the m-th risk unit.

2 . The method for evaluating the driving risk level in the tunnel based on the vehicle bus data according to claim 1 , wherein the step of extracting the driving risk characteristic feature indexes by using the sliding window method according to the bus data comprises: with T1 as a sliding time window, and T2 as a step size, extracting driving risk characteristic feature indexes of the five driving behavior features under each time window by using the sliding window method, and setting each time window as a driving risk unit to obtain the plurality of driving risk units, wherein each driving risk unit of the plurality of driving risk units comprises the 20 driving risk characteristic feature indexes.

3 . The method for evaluating the driving risk level in the tunnel based on the vehicle bus data according to claim 1 , wherein a method of training the LDA model comprises:
sampling all word sequences randomly to obtain a training set; constructing the LDA model and setting a number of model iterations; setting a series of risk theme numbers k, and dividing word sequences of driving risk units into k risk themes; for each risk theme number k, training the LDA model iteratively by using the training set to obtain a perplexity and a coherence score of a model corresponding to each risk theme number; and determining an optimal risk theme number k m of according to the perplexity and the coherence score.

4 . The method for evaluating the driving risk level in the tunnel based on the vehicle bus data according to claim 3 , wherein the risk level of each risk theme is determined by visually analyzing the driving risk characteristic feature indexes in each theme for k m risk themes.

5 . A system for evaluating a driving risk level in a tunnel based on vehicle bus data, comprising:
a data acquisition and analysis module, wherein the data acquisition and analysis module is configured to acquire bus data in a vehicle driving process in the tunnel, and extract driving risk characteristic feature indexes by using a sliding window method according to the bus data; wherein the bus data in the vehicle driving process comprises five driving behavior features of a steering wheel angle, a steering wheel angle change rate, an accelerator pedal position, a brake pedal position and a brake pedal change rate; each driving behavior feature has four driving risk characteristic feature indexes of an average value of the driving behavior feature within a time window, a standard deviation of the driving behavior feature within the time window, a maximum value of the driving behavior feature within the time window, and a minimum value of the driving behavior feature within the time window; a feature index encoding module, wherein the feature index encoding module is configured to write a feature index codebook according to the driving risk characteristic feature indexes, and obtain a word sequence under each time window according to the feature index codebook, which comprises: dividing values of the driving risk characteristic feature indexes into several ranges, so that each range corresponds to a word, which means encoding each driving risk characteristic feature index according to data distribution and domain knowledge to obtain the feature index codebook, wherein 20 driving risk characteristic feature indexes of each driving risk unit correspond to 20 words, respectively, and each driving risk unit is represented by a word sequence:

d
m

=

[

w

m
⁢
1

,

w

Assignees

• Zhejiang Univ Zhongyuan Institute
• Univ Zhejiang

Inventors

• Bai Congcong
• Jin Sheng
• Yang Chengcheng
• Zhong Saijun
• Zhou Mengtao
• Jiang Yang

Classifications

• B60W2540/18

  Steering angle · CPC title

• B60W2540/12

  Brake pedal position · CPC title

• B60W2540/10

  Accelerator pedal position · CPC title

• B60W50/0097

  Predicting future conditions · CPC title

• G06N7/01

  Probabilistic graphical models, e.g. probabilistic networks · CPC title