Driver Behavior Modeling Using Hybrid Dynamic Systems for ‘driver-aware’ Active Vehicle Safety

Modern safety systems are transforming vehicles from human-controlled passive devices into human-centric intelligent/ active systems. There is a wide range of systems from fully autonomous vehicles to humanaugmented control devices which have emerged in this field. In current trends, co-operative active systems have the driver in the decision and control processes are favored for their ‘human-centric’ approach. However, these systems pose a challenge in the design process since obtaining reliable human behavior models are difficult due to the complex nature of driving task in a dynamic traffic environment. From a control theory perspective, driving can be seen as a combination of continuous control segments combined with a discrete decision process. In this study, we will model driver behavior utilizing Hybrid Dynamic Systems (HDS) combining stochastic modeling tools (such as Hidden Markov Models) with control theoretic models. A subset of CAN-Bus and video channels from a demographically balanced UTDrive Corpus containing video (2 channels: driver and road scene), audio, and CAN-Bus signals of realistic driving sessions for 77 drivers are used to verify HDS models of lateral and longitudinal control behaviour. The model is used to suggest ‘driver-aware’ active safety system capable of assisting the driver in several lateral control tasks; lanekeeping, curve-negotiation and lane changing.