Anomaly Detection in Driving by Cluster Analysis Twice

TL;DR

This paper introduces ADDCAT, a novel anomaly detection method for driving that uses double clustering of sensor data to identify deviations from normal patterns without requiring prior training or high computational costs.

Contribution

The paper presents ADDCAT, a new unsupervised clustering-based approach for detecting driving anomalies efficiently without training.

Findings

Effective detection of driving anomalies demonstrated on open dataset

No prior training or extensive computation needed

Clusters accurately represent normal driving patterns

Abstract

Events deviating from normal traffic patterns in driving, anomalies, such as aggressive driving or bumpy roads, may harm delivery efficiency for transportation and logistics (T&L) business. Thus, detecting anomalies in driving is critical for the T&L industry. So far numerous researches have used vehicle sensor data to identify anomalies. Most previous works captured anomalies by using deep learning or machine learning algorithms, which require prior training processes and huge computational costs. This study proposes a method namely Anomaly Detection in Driving by Cluster Analysis Twice (ADDCAT) which clusters the processed sensor data in different physical properties. An event is said to be an anomaly if it never fits with the major cluster, which is considered as the pattern of normality in driving. This method provides a way to detect anomalies in driving with no prior training processes and huge computational costs needed. This paper validated the performance of the method on an open dataset.