Driving Behaviour Analysis Using CNN-LSTM Architectures and Temporal Feature Learning

Abstract

This paper presents a novel approach to driving style recognition, aiming to address the challenges of accuracy, speed, and adaptability in existing methods. By leveraging deep learning techniques, including Convolutional Neural Networks (CNNs) and Long Short-Term Memory (LSTM) networks, the proposed method achieves remarkable results in driving style classification. Initially, a method is devised to collect and preprocess driver operation time sequences, considering the limitations of available data types. Subsequently, CNNs are employed to extract essential features from the driving data. Additionally, LSTM modules are integrated to encode temporal information and enhance feature representation. The results demonstrate a significant improvement in both accuracy and speed, with driving style recognition accuracy surpassing 93%. Moreover, the utilization of CNNs and LSTM networks ensures robust performance in real-time scenarios. Furthermore, an ensemble approach combining CNN and LSTM models further enhances prediction accuracy, reaching up to 99%. As an extension, the study proposes exploring additional ensemble techniques, such as CNN + LSTM + Bidirectional LSTM, for even greater accuracy. Moreover, a user-friendly front-end interface is developed using the Flask framework, facilitating user testing and authentication. Overall, the proposed approach offers a highly accurate, fast, and adaptable solution for driving style recognition, with promising implications for real-world applications.