Abstract
This study is to develop a technique of behavior observation to detect insider□s abnormal behavior and workers’ emergent situation such as heart attack or fall down in NPP in order to prevent safety hazard due to workers’ unusual behavior. Moreover, experiment has been carried out in the viewpoint of human factors engineering for the purpose of comparing conventional surveillance system and the surveillance system with Unusual Behavior Detection support system (UBDSS) to measure system effectiveness and workload reductionFirst of all, the scenario and possible abnormal behavior in NPP were discussed. Next, the simulated video data was shot and, with image process algorithm, captured and transferred into three numerical variables that interpreting human action including gesture, moving speed and moving angle. The three variables were then analyzed with decision tree method to construct a model for finding relation among the three variables and the behavior observed. The rules of decision tree classification were used to predict new video data, which formed UBDSS. Afterward, an experiment was conducted to verify system effectiveness and reduction of workload by computing response time, error rate and NASA TLX task load index.The results of this study indicated that unusual behavior prediction rate of UBDSS is acceptably high (average classification rate = 83.01), and auto-alarm system with UBDSS is able to improve system performance as results of experiment indicated that the response time decreased about 42.9% and the error rate decreased about 62.1%. The result of this study can be applied on other relevant industries such as suicide preventions of labor-intensive industry and secure protection in safety-concerned facilities such as government office.