주거 및 공공장소 이상행동탐지를 위한 서비스 설계

초록

This study presents the construction of an AI learning dataset and the prototypical development of a model for detecting anomalous behaviors in residential and public spaces, as part of an effort to enhance security and crime prevention. The AI learning dataset and model were designed to stimulate the advancement of AI technology and the development of AI projects in private companies. During the dataset construction, ResNet50 was modularized to extract features from video frames, and 3D-CNN was used to process skeleton points from JSON files. This data was then labeled according to predefined anomalous behaviors. Furthermore, GridCV was employed to utilize the SVM classifier and GRUs for processing video sequences. The learning performance evaluation of the model demonstrated a continuous improvement in main accuracy and a decreasing trend in detailed loss.. The trained model can predict the category of behavior appearing in a given video sequence. The AI learning dataset and model prototyped in this study provide valuable insights into the practicality of developing anomaly detection functions. It is expected to contribute to the advancement of AI applications in the field of public safety by securing, constructing, and distributing data necessary for AI learning for immediate anomaly detection, crime prevention, and offender apprehension.

키워드