Audio based detection of rear approaching vehicles on a bicycle.

摘要

Cycling is an efficient mode of travel widely used for transport, recreation and sport all over the world. In addition to being environmentally friendly, it also affects the health of the cyclist favorably. Safety is an important concern for a cyclist because, during an accident with a motor vehicle, the cyclist is exposed to higher risk of injury than the vehicle driver. Improving bicycle safety is an important factor in saving lives and promoting the use of this environmentally friendly mode of transport.;The Cyber-Physical Bicycle system was introduced as a concept bicycle that can alert the cyclist of dangerously approaching vehicles from the rear. The system aims to accurately detect and track vehicles approaching from the rear, differentiate dangerously approaching vehicles, and alert the cyclist early enough to take preventive measures. This is achieved through video based detection. The traditional bicycle is extended with computational capabilities and a rear facing video camera, which constantly monitors vehicular traffic behind the bicycle. Research indicates that the system is feasible, works with good accuracy and generates timely alerts, though it cannot operate at full efficiency while running in realtime.;In this thesis, we present an approach that augments a bicycle with audio based detection of rear approaching vehicles. The audio based Cyber-Physical Bike continuously listens to the environment behind the bicycle with a microphone, detects rear approaching vehicles and alerts the biker to their presence. We describe the design for an audio based Cyber-Physical Bike and demonstrate its feasibility through evaluation of our prototype. We found that distinguishing the directionality of vehicle approach is a significant problem in case of audio, due to the similarity in sounds. Subsequently, we identified several audio features that help us differentiate rear and front approaching vehicles accurately. We also used a rear facing microphone to improve detection. Results show that our approach works with comparable accuracy to the video based approach, performs real time detection at lower energy and hardware costs, and is more efficient. However, the system sacrifices on timeliness of alerts, and the alerts are generated much later when compared to video based detection.