Developing a driver observation system requires concrete information of the driver's awareness and intentions. In a collision warning system, the warning signal should be adjusted based on the available information of the driver's state. In this scope, we develop new methods to estimate the awareness of the driver in car following and braking situations. We also present a novel method for detecting lane change based on wavelet decomposition of the lateral acceleration. Simulation results show that we can detect a lane change half a second earlier than conventional approaches.