Authors:
Matheus Silva de Almeida
;
Julia Ladeira
;
Caio Vicentin
;
Andre Costa
;
Marcia Pasin
and
Vinícius Marini
Affiliation:
Centro de Tecnologia, Universidade Federal de Santa Maria, Brazil
Keyword(s):
Efficient Processing, Dangerous Driving, Artificial Intelligence, Machine Learning Techniques.
Abstract:
Detection of dangerous driving behavior is a key component to improving road safety. It can be successfully carried out using data collected by sensors widely available in smartphones. Current work focuses on two groups: either they classify drivers in a binary way, into good and bad drivers, or they provide a scoring scale, allowing for a larger group of categories. This detection of dangerous driving behavior can be done with high granularity, evaluating a total distance covered by the driver on a trip, or with minute granularity, through the evaluation of small sections of driving, also making it possible to identify which maneuvers the driver is carrying out negligently. However, the process of collecting data for dangerous driving behavior is complicated because the driver needs to carry out these maneuvers, so that a classifier can later detect them, adding to situations of insecurity in traffic. Moreover, the solution needs to execute efficiently, so that the detection of dang
erous driving behavior can be carried out in real time. Given this problem, we propose a time efficient dangerous driving detection system using vehicle inertial data. In contrast to other works, we collected data in a simulation environment with a model car that allows us to perform risky maneuvers, which would not be possible in a real environment. We identify in our small dataset the dangerous driving behavior pattern. Thus, given the established pattern, we applied a machine learning method to generate a classifier to enable the detection of dangerous driving behavior. The resulting system achieved a total average accuracy of 85.61% in our experiments using a small dataset as input towards efficient data processing.
(More)