Vehicle Driving State Recognition and Test Analysis Using Vehicle Body Attitude Measurement and One-Dimensional Convolutional Neural Network

• DOI: 10.15282/ijame.22.2.2025.2.0939
• Publication Date: 2025-06-02T04:52:18Z
• Authors: Feifei Yu; Yishen Zhuo; Xiaoting Xing; Xiaoqing Yang; Guoyan Chen; Canyi Du

The increase in the number of cars on road has led to frequent traffic accidents, some which may be due human factors such as poor driving habits. In order reduce there is a need accurately analyze driver's behavior and identify actual state motion car. A method based one-dimensional convolutional neural network determine condition vehicle proposed. Vibration acceleration indications for four different fuel-powered traveling states (constant speed, acceleration, deceleration, deceleration) six electric driving, left-turn right-turn driving) were measured using GPS inertial navigation sensors. New data samples are then extracted evaluated cover full range each operational situation. Access "keras" package via Python allows creation Convolutional Neural Network (1D-CNN) model. model receives vibration signals input parameter tuning. experimental results show that application multimodal fusion automotive recognition achieved by combining attitude with 1D-CNN data. This approach features high accuracy, strong generalization ability, short training time, reliability, achieving accuracy 90%. It aimed at monitoring driver optimizing assistance systems, while also providing new ideas methods improving performance safety automatic autonomous systems.