Enhancing Pedestrian and Bicyclist Safety through Abnormal Driving Behavior Detection

Abnormal driving poses a significant risk not only to the driver (referred to as the ego vehicle) but also to other road users, particularly pedestrians and bicyclists. Existing literature (Wu et al., 2018) has shown that early detection and intervention in cases of abnormal driving can prevent traffic accidents or at least reduce their severity. To this end, this project is focused on creating an application designed to improve the safety of pedestrians and bicyclists by identifying abnormal driving behaviors.

Current research on detecting abnormal driving behaviors largely depends on the use of onboard sensors that monitor various aspects of the driver’s physical state and driving patterns, such as facial and gaze direction, blood pressure, and heart rate. However, this approach requires drivers to equip their vehicles with specific devices, often at their own expense, which may hinder the popularization of such technologies. An alternative and more cost-effective solution involves using roadside sensors, like cameras, LIDAR, and RADAR, to detect abnormal driving. This method analyzes and predicts vehicle trajectories based on live-streamed data from roadside sensors. If a vehicle’s trajectory significantly deviates from its predicted path, it can be identified as abnormal. Despite its promising results, this method has strict requirements regarding the accuracy of trajectory prediction. Otherwise, too many false alarms could erode public trust in the technology. Addressing these challenges and improving the reliability and accuracy of abnormal driving detection methods is a key goal of this project.