Real-Time Driving Distraction Recognition Through a Wrist-Mounted Accelerometer.

Area of Science:

• Human-Computer Interaction
• Transportation Safety
• Wearable Technology

Background:

• Distracted driving is a major cause of fatal vehicle accidents annually.
• Body-worn sensors offer a promising approach to mitigate driver distraction.
• Real-time detection of driver distraction can significantly enhance road safety.

Purpose of the Study:

• To develop and evaluate a real-time driver distraction recognition method.
• To investigate the effectiveness of a wrist-worn inertial measurement unit (IMU) for detecting manual distractions.
• To compare the performance of different deep neural network classifiers for identifying distracted driving behaviors.

Main Methods:

• Twenty participants engaged in simulated driving tasks involving common distracted behaviors.
• Acceleration data from wrist-worn IMUs were analyzed using 2-second data streams.
• Three deep neural network classifiers (ConvLSTM, CNN, LSTM) were trained and evaluated using F1-scores.

Main Results:

• The convolutional long short-term memory (ConvLSTM) network demonstrated superior performance in recognizing distracted driving behaviors.
• Within-participant F1-scores were 0.87 for ConvLSTM, 0.82 for CNN, and 0.82 for LSTM.
• Between-participant F1-scores were 0.87 for ConvLSTM, 0.76 for CNN, and 0.85 for LSTM.

Conclusions:

• The proposed system utilizing a wrist-worn IMU and ConvLSTM classifier shows potential for real-time driver distraction mitigation.
• This approach may contribute to improving overall transportation safety.
• Further development of this pilot study could lead to practical applications in driver safety.