Cost-effective AIoT-based hazard detection using 2D camera in manufacturing environments

In industrial environments, forklift safety management systems using various smart sensors such as light detection and ranging (LiDAR), laser, ultrasound, radio frequency identification (RFID) sensors, or much more have been developed to prevent forklift accidents arising from collisions. However, these approaches tend to give a big burden to the introduction of the forklift safety management system because they typically require the high initial costs for its implementation and the complexity of the system installation, configuration, and maintenance. Hence, in this study, the authors propose a simple, cost-effective, and straightforward mechanism and system architecture based on the artificial intelligence of things (AIoT). The proposed system and algorithm have been designed for the early detection of potential hazards and dangerous environments involving moving forklifts and pedestrians. In this paper, we describe a prototype implementation of AIoT-based forklift safety management system and its performance evaluation results. From the laboratory and field tests, it was shown that the spatial-perspective-based algorithm proposed in this study improves the F1 score value, which indicates the critical metric for safety assessment, by approximately 26% and 2.4% in the laboratory and the field test, respectively, when compared to the distance-based algorithm only based on the calculated distance. It was analyzed that this significant discrepancy in the F1 score's performance improvement between the laboratory and the field test was attributed to the frequency of scenes with pedestrians near forklifts. Consequently, the field test demonstrated that the cost-effective 2D camera and AIoT-based hazard detection using the proposed algorithm has effectively and correctly identified various hazardous situations that happen in the real manufacturing workspaces, including the scenarios that the pedestrians are partially detected in the images while approaching the forklift so closely.