Method and system for detecting driving anomalies

Patent metadata

Field	Value
Publication number	US-11807253-B2
Application number	US-202117476061-A
Country	US
Kind code	B2
Filing date	Sep 15, 2021
Priority date	Sep 17, 2020
Publication date	Nov 7, 2023
Grant date	Nov 7, 2023

Abstract

According to one embodiment of the invention, a method for detecting driving anomalies comprises steps of: with at least one algorithm, processing raw data from On-Board Diagnostics of a car to generate time-series data, with the time series data as input, using an automatic driving behavior separating technology to identify a plurality of driving behaviors; with the driving behaviors as input, using an artificial intelligence technology to build up a driving anomaly detection model without labeling the driving behaviors; and issuing an alarm for driving anomaly identified according to an analyzing result of alarm signature of the driving behaviors.

First claim

What is claimed is:

1. A method for detecting driving anomalies, applied in a system for detecting driving anomalies comprising a wireless communication module and a micro-processor, the method for detecting driving anomalies, comprising steps of:
with at least one algorithm, with the micro-processor, pre-processing raw data from On-Board Diagnostics of a car to generate time-series data;
with the time series data as input, with the micro-processor, using an automatic driving behavior separating technology to identify a plurality of driving behaviors of the car;
with the driving behaviors as input, with the micro-processor, using an artificial intelligence (AI) technology to build up a driving anomaly detection model to detect driving behaviors of the car without labeling the driving behaviors;
with the micro-processor, issuing an alarm for driving anomaly identified when at least one statistic feature exceeds a predetermined threshold or distributes abnormally in a predetermined time range according to an analyzing result of alarm signatures of the driving behaviors analyzed with:
setting a detecting window in the form of one of sliding window and fixed window; and
calculating the at least one statistic feature of the alarm signatures of the driving behaviors in the detecting window.

2. The method for detecting driving anomalies according to claim 1 , wherein the step of pre-processing raw data with at least one algorithm comprises a step of performing classification, clustering, or dimension deduction on the raw data sampled at the same time to generate the time series data.

3. The method for detecting driving anomalies according to claim 1 , wherein at least one algorithm comprises one of Gaussian mixture model algorithm (GMM algorithm), K-means algorithm, and principal component analysis (PCA).

4. The method for detecting driving anomalies according to claim 1 , wherein the automatic driving behavior separating technology utilizes natural language processing (NLP).

5. The method for detecting driving anomalies according to claim 1 , wherein the AI technology comprises one of machine learning algorithm and deep learning algorithm, the machine learning algorithm comprises one of supervised machine learning algorithm and unsupervised machine learning algorithm, and the deep learning algorithm comprises one of convolution neural networks (CNN) and recurrent neural network (RNN).

6. The method for detecting driving anomalies according to claim 1 , wherein at least one statistic feature comprises one of:
An expectation E is calculated by: (the number of abnormal alarm signatures in the detecting window)/(the number of the alarm signatures in the detecting window);
a joint Shannon entropy of co-occurrence matrix

∑

i
=
0

1

⁢

∑

j
=
0

1

⁢

P

i
,
j

*

log
⁡

(

P

i
,
j

)

,

 wherein P i,j =P[X t−1 =i, X t =j]; and
a Kullback-Leibler divergence between the alarm signatures in the detecting window and alarm signatures in training data

D
KL

⁡

(

P
⁢




⁢
Q

)

=

∑
i

⁢

P
⁡

(
i
)

⁢
log
⁢

P
⁡

(
i
)

Q
⁡

(
i
)

,

 wherein P and Q are distribution functions of the alarm signatures in the detecting window and the alarm signatures in the training data, respectively.

Assignees

• Univ Nat Taiwan

Inventors

• Lin Phone
• Lin Xin-Xue
• Yeh En-Hau
• Lee Chia-Peng

Classifications

• G06N3/0442

  characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU] · CPC title

• G06N3/0464

  Convolutional networks [CNN, ConvNet] · CPC title

• G06N3/09

  Supervised learning · CPC title

• B60W50/0205 Primary

  Diagnosing or detecting failures; Failure detection models · CPC title

• B60W50/14

  Means for informing the driver, warning the driver or prompting a driver intervention · CPC title