CHARM: Context-aware Human Activity Recognition and Monitoring
for Intelligent Vehicles

Today’s automobile is nearly autonomous due to Advanced Driver-Assistance Systems (ADAS) ranging from adaptive cruise control, automatic emergency braking, auto-parallel parking to lane-keeping assistance and automated steering. This is feasible due to ultrasonic sensors, cameras, radars and lidars, focusing on surrounding environmental perception. However, minimal work has focused on the human driver perspectives and human error is overwhelmingly to blame for most accidents. The role of the driver could be taken over by automation, but the vehicle also requires delivering performance identical to that of a driver if it is to be trusted. As a result, this automation could lead to overreliance on incomplete ADAS. Therefore, the ADAS must focus on understanding, modelling and predicting human agents, as well as on the surrounding traffic conditions, making it a key component of knowing how vehicles will learn to adapt to various driving conditions and environments.

CHARM’s aim to develop a novel context-aware computational framework for monitoring and anticipating human driver’s activity (distractions, eating, drinking, etc) in a given context (congestion, weather, road works, traffic sign, etc). This is a difficult problem involving computer vision (CV), big data and artificial intelligence (AI), with immediate applications to smart cyber physical systems/environments. The framework is targeted for the ADAS for safe driving, Take-Over-Request (TOR) for the next generation intelligent autonomous vehicles, as well as more natural human-car/robot/machine interactions associated with the UKIERI-DST cyber physical systems discipline.