Abstract
Currently, Thailand has a higher death rate from motorcycle accidents than from other vehicles. Wearing a helmet can reduce the severity of head injuries in accidents, since the head is a crucial part of the human body. The objective of this research was to evaluate the effectiveness of AI camera detection of helmet-wearing behaviours (both drivers and passengers) and to find the relationship between the accuracy of AI cameras in detecting helmet-wearing behaviours and the camera installation points. Data was collected from two AI camera installation points, observing a total of 674 motorcycles. The data was analyzed using descriptive statistics and the Pearson Chi-Square test, as well as a logistic regression analysis. The study found that AI cameras could accurately detect helmet-wearing behaviours, with Point 1 having a detection accuracy of 73.7% and Point 2 having a detection accuracy of 89.5% (Point 1 was a two-lane, one-way roadway featuring a covered pedestrian walkway with high foot traffic; whereas Point 2 was a two-lane, bi-directional roadway flanked by drainage ditches on both sides with no dedicated pedestrian facilities.). Additionally, there was a significant difference in the detection accuracy of the AI cameras between the two installation points (p-value < 0.05), as well as a difference in the detection accuracy based on helmet-wearing behaviours (p-value < 0.05). In addition, analysis revealed that camera location and helmet type had significantly impacted detection accuracy. Specifically, Point 2 outperformed Point 1, and Type 2 helmet detection showed a 23-fold higher accuracy compared to other types (p-value < 0.05)