Cheung, Chi-Chung (The Hong Kong Polytechnic University), Cheung, Chun-Ming (The Hong Kong Polytechnic University), Tam, Pak-Sam (The Hong Kong Polytechnic University), Cheung, Ki (The Hong Kong Polytechnic University), Wong, Hei-Nam (The Hong Kong Polytechnic University), Hui, Siu-Ki (Howin Development Company Limited), Tong, Chi-Wang (Howin Development Company Limited)

DNN Classifiers with Segmentation Method and Indirect Labeling Mechanism to Classify Driving Behaviors through a Smartphone-Based IoT System

Scheduled for presentation during the Poster Session "Travel Behavior Under ITS" (WeAT16), Wednesday, September 25, 2024, 10:30−12:30, Foyer

2024 IEEE 27th International Conference on Intelligent Transportation Systems (ITSC), September 24- 27, 2024, Edmonton, Canada

This information is tentative and subject to change. Compiled on October 3, 2024

Abstract

Traffic accidents are one of the most common problems in the world, causing significant casualties and property losses. There are different ways to reduce the number of traffic accidents. One of them is to identify bad driving behaviors and give recommendations or advice according to the bad driving behaviors. Some research efforts have been made to classify driving behaviors. For the simplicity of the implementation, most of them install a smartphone in a car and measure the movement of the car. Then, they apply different algorithms, such as some simple inequalities or machine learning algorithms, to classify the driving behaviors of the driver in the car. However, most of them cannot be applied in real scenarios. Moreover, their databases are usually small because it is difficult to do the labeling. This paper proposes a segmentation method to extract events (segments) from the raw data and build a database in real scenarios. Moreover, we propose an indirect labeling mechanism to effectively label events to build an extensive database. We develop DNN (Deep Neural Network) classifiers to classify the driving behaviors in this database. Through our performance investigation, we can properly classify driving behaviors in real scenarios.