정규화 기법을 통한 안면 인식 알고리즘 성능 향상에 관한 연구

Abstract

Through the combination of computer vision technology and artificial intelligence, facial recognition technology is drawing attention as a new means of personal authentication in the era of the fourth industry. Facial recognition technology uses imaging equipment to photograph a person's face and extract characteristic data. The extracted data are matched against the facial features of the stored database. Facial recognition technology is a contactless technology compared to other biometric recognition technologies, which is used in various fields due to its high hygiene, convenience and security, and in particular, safety accidents in workplaces are closely related to life, and various studies related to workplace safety management using intelligent video information are being conducted in the manufacturing industry. In this paper, a study is conducted on the development of facial recognition algorithm using deep learning to control worker access in hazardous areas. The accuracy of the recognition of the proposed facial recognition algorithm (object detection algorithm (SSD) and object recognition algorithm (ResNet)) is closely related to the safety of the operator. Therefore, the goal is to analyze the relationship between various normalization techniques (Min-Max Scaler, MaxAbs Scaler, Standard Scaler) and the recognition rate of the proposed facial recognition algorithm to propose a high-accuracy facial recognition algorithm. In the future, we will conduct research on safety issues in the manufacturing industry based on facial recognition and image recognition technologies.