Signature-Based Pattern Recognition: Application to Closed-Loop Driving Maneuvers

In order to detect specific steering maneuvers in closed-loop driving evaluations on open road, a novel pattern recognition approach is proposed. Based on signature analysis, this approach incorporates both signal processing techniques and expert knowledge. Key properties of this approach are the ability to adapt to the complexity of patterns it recognizes and its good sensitivity in distinguishing similar patterns. The objective of this paper is two-fold. First, a novel pattern recognition algorithm is introduced, and its performance is assessed in terms of "miss" and "false positive" rates. Second, different optimization algorithms were compared, which were used to improve recognition accuracy by finding an optimal set of classification features. This paper describes the approach which is suitable to recognize a wide range of patterns in different classes of signals. The approach was validated in the domain of automotive engineering, but it is generic enough to be applied to other domains where instrumented tests and measurements are commonplace.