Seen and missed traffic objects: A traffic object-specific awareness estimation

Abstract

Handing-over vehicle control from a human driver to an intelligent vehicle and vice versa needs elaborate and safe hand-over strategies. Before passing control it must be ensured that the driver is aware of all objects which are important in a particular traffic situation. In this work a decision tree is used to learn which objects attract the driver's gaze in a particular situation. The decision tree classifies on object features as the object's type, velocity, size, color, and brightness. This information is fused from laser-scanners, front camera, and the vehicle's CAN-bus data. Whilst driving, an awareness confidence is built for each object perceived by the laser-scanners. Unexpected gaze behavior is detected by comparing the awareness confidence of each object to the expected gaze behavior, learned by means of the decision tree. Objects overlooked by the driver are further classified as critical or uncritical. This provides valuable information for following human-car interaction, augmented-reality, or safety applications.