비정상적인 상황 검출을 위한 임베디드 시스템 구현에 관한 연구

Abstract

The conventional CCTV surveillance system for preventing accidents and incidents is a method for surveillance of multiple monitors by surveillance personnel. However, where one person monitors multiple CCTVs, the person misses 95% of the data after 22 minutes. To address this issue, researchers have studied the computer-based intelligent image surveillance system for notifying people of abnormal situation in the images. However, because the system is involved in the issues of power consumption and costs for building the system, and privacy leaks, the intelligent image surveillance system based on embedded modules has been studied. This thesis implements the intelligent image surveillance system based on embedded modules for detecting intruders on the basis of learnt information, detecting fires on the basis of colors and motion information, and detecting loitering and fall down objects on the basis of human body motion information. Moreover, the algorithm and the embedded module optimization method are applied to implement real-time processing. The method for optimizing the used algorithm is used for 6 processes of examining subroutines, using positive integers, minimizing remainders and use of division operators, minimizing direct calculation of global variables, minimizing transfer factors and recursive algorithms in function call, and using Cython. The method for optimizing embedded modules is used for 3 processes of removing bottlenecks, using memories, and using zRam. The intelligent image surveillance system based on embedded modules is implemented in Raspberry Pi. The algorithm processing time is 0.053 seconds on the average by a computer, 0.9499 seconds on the average by Raspberry Pi before optimization, and 0.47 seconds on the average by Raspberry Pi after optimization. The result of comparing before and after optimizing the intelligent image surveillance system shows reduced processing time by 50.52%. Therefore, this suggests real processing of the intelligent image surveillance system based on the embedded modules is possible.