Towards a Practical Pedestrian Distraction Detection Framework using Wearables

Abstract

Pedestrian safety continues to be a significant con- cern in urban communities with distraction being one of the main contributing factor behind serious accidents involving pedestrians. The advent of sophisticated mobile and wearable devices, equipped with high-precision on-board sensors capable of measuring fine-grained user movements and context, provides a tremendous opportunity for designing effective pedestrian safety systems and applications. Accurate recognition of pedestrian distractions in real-time given the memory, computation and com- munication limitations of these devices, however, remains a key technical challenge in the design of such systems. Earlier research efforts in this direction have primarily focused on achieving high distraction detection accuracy, resulting in techniques that are either resource intensive and unsuitable for implementation on mainstream mobile devices, or computationally slow and not real- time, or require specialized hardware and thus less likely to be adopted by most users. Our goal in this paper is to design a pedestrian distraction detection technique that overcomes some of these shortcomings (of existing techniques) and achieves a favorable balance between computational efficiency, detection accuracy, and energy consumption.