Continual Learning for Behavior-based Driver Identification

TL;DR

This study evaluates continual learning for behavior-based driver identification to address real-world deployment constraints, including limited compute, adaptation to new drivers, and evolving driver behavior. It compares standard CL approaches (ER, DER, EWC, LwF) on the OCSLab dataset across three progressively realistic scenarios and introduces two novel prediction-smoothing methods, SmooER and SmooDER, that exploit temporal continuity. The results show that DER-based methods can approach near-static performance (about an 11% drop), while SmooDER reduces the loss to roughly 2–3%, demonstrating the practicality of CL for in-vehicle deployment. The work also analyzes memory and time efficiency, showing that 12 MB of memory suffices for high-performance continual learning, enabling on-cloud or direct-vehicle deployment, and provides open-source implementations to promote reproducibility and future research.

Abstract

Behavior-based Driver Identification is an emerging technology that recognizes drivers based on their unique driving behaviors, offering important applications such as vehicle theft prevention and personalized driving experiences. However, most studies fail to account for the real-world challenges of deploying Deep Learning models within vehicles. These challenges include operating under limited computational resources, adapting to new drivers, and changes in driving behavior over time. The objective of this study is to evaluate if Continual Learning (CL) is well-suited to address these challenges, as it enables models to retain previously learned knowledge while continually adapting with minimal computational overhead and resource requirements. We tested several CL techniques across three scenarios of increasing complexity based on the well-known OCSLab dataset. This work provides an important step forward in scalable driver identification solutions, demonstrating that CL approaches, such as DER, can obtain strong performance, with only an 11% reduction in accuracy compared to the static scenario. Furthermore, to enhance the performance, we propose two new methods, SmooER and SmooDER, that leverage the temporal continuity of driver identity over time to enhance classification accuracy. Our novel method, SmooDER, achieves optimal results with only a 2% reduction compared to the 11\% of the DER approach. In conclusion, this study proves the feasibility of CL approaches to address the challenges of Driver Identification in dynamic environments, making them suitable for deployment on cloud infrastructure or directly within vehicles.

Figures (8)

• Figure 1: General representation of the experimental setup. Since the goal is to learn new tasks as they arrive while retaining knowledge from previous ones, the test set for each task consists of all the test data from drivers whose labels already appeared.
• Figure 2: Task sequence representation for Scenario 1. Given ten drivers, each task contains information about two new drivers, for a total of five tasks.
• Figure 3: Task sequence representation for Scenario 2. A single driver is added to the system each time (except for task 1).
• Figure 4: Task sequence representation for Scenario 3. The data is divided into drive sessions (there are two of them for each driver) and each task contains the data of two sessions. So the next task can either introduce new drivers or expand the knowledge of previously seen drivers.
• Figure 5: Results for Scenario 1.