Leveraging sensor information from portable devices towards automatic driving maneuver recognition

With the proliferation of smart portable devices, more people have started using them within the vehicular environment while driving. Although these smart devices provide a variety of useful information, using them while driving significantly affects the driver's attention towards the road. This can in turn cause driver distraction and lead to increased risk of crashes. On the positive side, these devices are equipped with powerful sensors which can be effectively utilized towards driver behavior analysis and safety. This study evaluates the effectiveness of portable sensor information in driver assistance systems. Available signals from the CAN-bus are compared with those extracted from an off-the-shelf portable device for recognizing patterns in driving sub-tasks and maneuvers. Through our analysis, a qualitative feature set is identified with which portable devices could be employed to prune the search space in recognizing driving maneuvers and possible instances of driver distraction. An absolute improvement of 15% is achieved with portable sensor information compared to CAN-bus signals, which motivates further study of portable devices to build driver behavior models for driver assistance systems.