MEKON 2021, İstanbul, Turkey, 25 June 2021, pp.1-6
A significant number of fatal traffic accidents with losses of lives and property all over the world are caused by driver errors. In this paper, a novel driver, vehicle, and environment monitoring system for traffic accident avoiding is presented. Our proposed monitoring system computes a single index, we call Driver Proficiency Index (DPI) using PERCLOS, SDLP, and TPM parameters in real-time to aid an accident avoiding system making decisions to take control of the vehicle from the driver. Data collected from cameras, biosensors, odometry fused, and an AI-based recommendation to reduce the risk of traffic accidents. The proposed solution has the ability not only to make real-time safe driving recommendations to help to reduce traffic accidents but also to build time history models that may help to compute the driver risk factor for the auto insurance industry. Based on some statistical information from previous research on the effects of the selected parameters on fatigue and inattentiveness, a decision tree model on the level of fatigue of the driver was created. In the study, real-time lateral position values using CARLA, biomathematical parameter values were simulated using FIPS. A three-layer NN is used to fuse all our parameters to generate a more robust continuous-time DPI signal. Data to train the ANN is obtained by various synthetic data generated using designed driving scenarios in our simulation platform. We have shown that using a more effective set of sensors a driver’s health, attentiveness and fatigue may be monitored in real-time better. As a result more causes of driver errors with high potential of traffic accidents are responded effectively.