BROWSE
Detail
Conference Paper
Abnormal Detection of Worker by Interaction Analysis of Accident-Causing Objects
Cited 0 time in 
Share 
Share
- Authors
- Won Shik Kim, Kye Kyung Kim
- Issue Date
- 2023-08
- Citation
- International Conference on Robot and Human Interactive Communication (RO-MAN) 2023, pp.1-6
- Language
- English
- Type
- Conference Paper
- Abstract
- In industrial sites, many industrial accidents cause human casualties every year, and deep learning-based object detection and danger zone management technologies are being proposed to minimize accidents. However, existing studies have focused only on object detection which has poor performance when a dangerous situation occurs by interacting two or more accident-causing objects. This paper proposes a method that detects accident risks in advance through object detection and risk interaction analysis between objects. It consists of four modules: image acquisition, object detection, worker action analysis, and risk event detection by dangerous object interaction. YOLOv4 is selected and fine-tuned to detect workers and conveyor objects that cause accidents by interacting objects. After selecting the danger and caution zones, it determines whether or not the detection object exists around the danger zone. A total of 68,621 image datasets collected from industrial sites were created to train and evaluate the system. The mAP of 91.79% for object detection is obtained and the F1 score of 88.9% for risk event detection is obtained.
- KSP Keywords
- Danger Zone, Event detection, Human casualties, Image datasets, Industrial accidents, Learning-based, Object detection, Object interaction, abnormal detection, deep learning(DL), image acquisition