TAMS: A CNN-based time attention network for time series sensor data with feature points of bicycle accident

Abstract

With Personal Mobility Vehicles (PMV) such as bicycles and electric scooters becoming a major means of transportation and delivery, the need to reduce injuries from accidents, which are also increasing, has become important. This study proposes a deep learning architecture called TAMS (Time Attention for Multi Sensor) based on Convolutional Neural Network (CNN) using Inertial Measurement Unit (IMU) sensor data. Through an evaluation and comparison with various deep learning algorithms, including CNN, Recurrent Neural Network (RNN), and Long Short-Term Memory (LSTM) networks, it shows that TAMS returns better accuracy and efficiency in real-time accident detection. Effectiveness was validated through experiments using a mannequin equipped with sensors, and a deep learning model was implemented on a Raspberry Pi to perform immediate accident detection and airbag deployment. This study contributes to improving the safety of PMV riders and lays the foundation for expansion to various types of PMVs beyond bicycles.