Fuzzy-based driver monitoring system to assess dangerous driving on roads

Road accidents result from various contributing factors, including driver fatigue, inappropriate vehicle speed, adverse weather, and temporal factors. The research in this paper aims to design and evaluate a Fuzzy Driver Monitoring System (FDMS) that automatically identifies dangerous driving behavior by considering critical driving parameters to enhance road safety.

In this work, a fuzzy logic driver alert system is designed that considers five key driving parameters: vehicle speed, driver drowsiness, weather, day of the week, and time of day. To detect the driver's drowsiness, a BlazeFace-based detection stage is first utilized to accurately identify and crop the driver's face from the video feed to ensure the model focuses on pertinent facial cues. The drowsiness level is then estimated using an improved deep-learning model (LSTM, CNN) with a longer temporal window for facial expression recognition. The FDMS evaluates driving risks from very low to very high according to its five inputs.

The system proposed here accurately evaluated driving risk levels under various simulated conditions. Scenarios involving high drowsiness of the driver, higher vehicle speeds, and poor weather conditions all yielded stable high-risk levels. Specifically, the improved drowsiness detection algorithm reached an accuracy rate of 70.46%, enhancing the reliability of risk assessment by including a broader range of risk factors than earlier studies. Furthermore, the model demonstrated a robust classification performance with an F1-score of 71.64% and an Area Under the Curve (AUC) of 0.75, confirming its effectiveness in distinguishing between drowsy and alert states. Additionally, a Graphical User Interface (GUI) was developed to display real-time data and the driving risk level based on simulated or collected data from the Global Positioning System (GPS) sensor, weather Application Programming Interface (API), and camera. The proposed FDMS was evaluated under various driving conditions and achieved an accuracy of 77.5% in true alerts provided to the driver. Finally, the proposed FDMS is experimentally assessed using an experimental hardware setup consisting of a laptop, webcam, GPS, and General Packet Radio Service (GPRS) module to demonstrate its real-world applicability.

The proposed FDMS is shown to detect high-risk driving conditions precisely with timely and precise risk estimation. The addition of various significant risk factors significantly enhanced prediction accuracy, indicating its potential for preventing many accidents through timely warnings to the driver.