Decision Fusion Techniques for In-Car Driver Recognition

In this study, we study various classifier fusion techniques to combine driver identity decisions coming from independent classifiers on different channels. Using trainable combining methods and driving behavior signals only, we find that the identification error can be reduced significantly. We also carry out a comparative study on the individual vs. combined performance of various driving behavior signals and observe that some driving signals carry more biometric information than others. We conclude that individual driving signals are not largely indicative of the person by themselves but combined multi-modal signal or classifier methods seem to be very successful in biometric identification in a car environment.