Human Body Detection In The Scene Of Disasters

Although the science and technology have developed with astonishing speed, for the arrival of disasters, we cann't forecast all of them accurately. Earthquakes, tsunamis, fires and other natural disasters are still raging swallowing countless creatures. After the disasters, how to relieve the victims of the disasters effectively has become the focus of the recent research. At present, the relief after disasters depends on human mostly. In the hazardous environments of the disaster site, sometimes the victims are not rescued, and due to the circumstances mutation, people who rescue victims may be faced with threaten. In this case, the use of the robots are imperative.This paper studies the context of the vision of the robots who relieve the victims. Mainly at the complex background of the disaster sites, we found people's physical characteristic information, and consequently by the orientation of the human body, we can determine the objectives which the robots will rescue. For the researches of the detection of the human body, domestic and foreign scholars have done a lot more in-depth work. However, for the background of this unique application, in the complex environment of the disaster site, the conditions of the block is very serious. And then most of the Methods offered by existing references of human body detection can not complete the task of identifying the human body. This paper uses a detection method based on local physical, dividing human body into face, head and shoulders, upper body, legs and feet and some other parts. First of all search the splitting local characteristics of the human body, and then using the search results of each detector, carry out integration. Finally analyze the existence of human by the local characteristics information and get the location of the human body. In this paper, we detecte the characteristics of the upper body and the human face using the Haar features, extract the corresponding features of the region and integrate the region with the the candidate regions which is detected by the skin color detector and the candidate regions of the head and trunk which is detected by the HOG features. Using the features of the multiple characteristics, compensate mutually and eliminate mistakes. Thereby we can get better detection effects. In the paper, we use DBSCAN clustering algorithm on fusing local test results get good results. In the end, we present the concept of features relationships, formulating a series of rules, and get many good effects from the experiments of fusing the results by using features relationships.It can eliminate the false detection of local detectors effectively, which makes detection results have been tremendous improved.