Abstract

Automated driving systems increase safety by providing stability assistance and collision warnings to drivers to avert traffic dangers. Various techniques including artificial intelligence and machine learning may be integrated for increased user benefit such as prediction of future driving scenarios including user actions and providing human-like warning interfaces. The intended action of a user or driver is a state of mind that is not measurable. However, input measurements from the vehicle’s trajectory and controls such as steering and angle are effective in indicating driver’s intention a few seconds in advance using machine learning. Driver actions that can be predicted include, intention to change lanes, overtake, turn etc. Many studies have developed techniques to predict driver intentions but have not significantly integrated it into an automated driving system to assist users. Whereas this enhancement is necessary as input to determine when to warn drivers about imminent traffic dangers and reduce false alarms. Thus, it has not been determined whether warning drivers based on their intended lane change action would further enhance their performance. In this study, a machine learning algorithm specifically random forest was used to predict when a driver intends to make a lane change which was then applied to generate warnings for the user to avoid imminent traffic collision. Three driving actions specifically, lane change to left and right and lane keeping intentions were predicted and utilised to warn drivers on a three-lane highway of potential collision. Using an audio-visual interface, six warnings corresponding to the six surrounding vehicle positions relative the subject driver were designed to indicate potential collision in the direction of the intended action when other vehicles are in close proximity. The results indicate reduced false alarms and risk of collision in relation to front vehicles when the predicted action is a lane change to left and to front-right and back-right vehicles when the predicted action is a lane change to right. Thus, the results of this study indicate that warning drivers against an intended action that might result in collision is useful especially in the case of a lane change to left.

Copyright of the paper named above is hereby assigned and transferred to the Nigerian Journal of Applied Science and Innovative Technology published by College of Engineering, Bells University of Technology, Nigeria.