Development of the Artificial Intelligence for Object Detection with YOLOv8 for Increasing Efficiency of Incidents Detection of Warehouse Forklift Operators in Rayong Province

Abstract

Detection and reporting incidents by employees are significant issues: employees still lack incident reporting. This study aims to (1) enhance the image processing capabilities of the YOLOv8 program using Raspberry Pi 5 to detect undesirable incidents in forklift operations and (2) compare the number of recorded incidents before and after implementing the improved YOLOv8 program. The improvement process includes (1) inputting the dataset, (2) training the model with two sets: one for distances inside one meter and another for distances between one and three meters, and (3) applying a 15% sampling rate. The research found that the AI-based image processing technique had an average object detection accuracy (mean Average Precision: mAP) of 85.35%. A study of forklift operator events found a statistically significant difference in the number of incidents before and after implementing AI-based incident detection, analyzed using the Shapiro-Wilk test at a significance level of p-value < 0.05. Furthermore, post-implementation user satisfaction was rated at the highest level, particularly in terms of increased benefits, compatibility, and the overall effectiveness of the YOLOv8 program.