A Warning System Based On Fatigue Detection And Moving Object Recognition

In recent years, as China’s rapid economic development and the increasingly serioussecurity situation at home and abroad, military and national security authorities for variousaspects of the security system demand is rapidly increasing. Actually, various accidents thatthe monitoring personnel of the forces and the national security sector are on fatigued dutycaused endless every year. Fatigue has also been the three major causes of various types ofaccidents in mind of the associated security sector, which becomes the main obstacle of itsprotect people’s lives and property safety. Therefore, in fact, how to take effective measures toprevent or find timely the situation of the monitoring personnel’s fatigue has great significancefor protecting the life and property safety and reducing the occurrence of accidents. On theother hand, since the special nature of the sectorial mandate, it is very important of itsperimeter security. In the real work, this aspect of the security is achieved by the people onduty use the video surveillance and observation. Therefore it need a very intelligent videosurveillance software that can automatically determine whether someone make a perimeterintrusion, to protect its perimeter security.In this system, we use a camera to take the real-time facial image capture for someone onduty. By processing and analyzing degree of closure of the human eye, we can determine thefatigue attendant to make timely warning. PERCLOS is our main evaluation criteria, which isthat how much percentage of total time does the period in which human eye is closed take.This paper introduces an improved algorithm based on AdaBoost cascade classifier with Haarfeature. Simultaneously, we detect the face in the target image at first, and then, within thedivision of the face image, we detect the human eye zone in the face image detection, and next,within the region in the human eye, we make the vertical integral projection. According to therelevant characteristics of the projection curve, we can determine the human eye closed state.Finally, by a certain criterion given in advance, we can determine whether a state of the eyefatigue, so as to achieve the purpose of fatigue testing. When someone is detected in fatiguestate, the system will automatically alarm to make him/her returne to normal, and thus try toavoid the security risks of the duty.The core algorithms in the system are an improved algorithm based on AdaBoost cascadeclassifier with Haar features and a vertical integral projection. In this paper, it introduces themain ideas of these two algorithms and working principle at first. And then, the characteristicsof actual working environment for staff on duty were analyzed and requirements analysis isgiven. Next, we achieve the entire design by using the OpenCV development kit which is a image processing kit developed by Intel. Finally, we test the system with the collected set ofhuman faces and record results. According to the result, we know that the proposed algorithmsare high accuracy, speed and strong robustness in face detection and determination of the stateof the human eye. At the same time, we analyze the deficiencies in the experiment, make theimproved method and prospect system development.