Analysis for Automatic Extraction of Uncomfortable Driving by Constant Observation of Route Buses Using Anonymous Sensors

Bus accidents caused by aging and understaffed bus drivers have become a significant social issue, and the high rate of in-vehicle accidents is particularly problematic for route buses. It’s crucial to reduce the number of these accidents. In this study, we created training data to automatically identify unpleasant driving conditions using sensor data collected on board route buses. To address privacy concerns and data volume, we utilized a low-resolution far-infrared camera for visual inspections inside the bus. Additionally, we analyzed vehicle speed, GPS data, and acceleration sensor data. Manual scrutiny of large sensor data volumes is time-consuming, so we focused on specific data points and approximate threshold values of acceleration based on the sensor characteristics and prior research. Through this approach, we selected suitable data refinement conditions from the cases identified and broadened the analysis scope to generate training data. Consequently, we found that gathering training data while reducing the processing time through the use of appropriate data refinement conditions was feasible.