Human Body Detect Technique Research On Disaster Site

Natural calamities take place frequency in recenlty years.After chinese Wen Chuan take place of heavy earthquake, Haiti,yueshu and chile immediately take place. The key problem of disaster site search and rescue is that people doesn't acquaint with the surrounding environment of the disaster area. Search and rescue person doen't clear know how much people were bury and It's location.The people who bury deeply in the reinforced concrete building was very difficulty to search and rescue. Rescue people in the narrow and small space is no doubt and biggest difficulty. Rescue int the characteristic environments,such as poisonous gas leakage and fire disaster is not very easy. Our study topic proposes to come under this back ground.This paper mianly studied the human body detect in the disaster site.infrared camera and optics camera was installed in the AS-RF robot.Use the ICON in the disaster site to judge whether existence life body. If include the live life we must return the location. This is advantageous to the search and rescue person.The human body detect back ground in this paper is more complicated,carriage is diverse. The place illumination is darker or didn't have illumiantion,we use the whole human body Character to detect infrared ICON. The ilumination on the suitable place,we use infrared ICON and optics ICON combine together to detect human body.First we must registration the infrared ICON and optics ICON to get a human body candidate for detect.we adoption detect the parts of bodies and get a result.we fusion this result to judge whether existence a person.This paper divided human body to shoulder(front,side),face,leg,skin color and hair five parts.Each part select respectively character. Use correspond method to train classifier. For mass experiment,the human body carriage in the disaster site pass the revolution of a certain angle will change into three kinds of classfier:Pedestrian human body model,sit human body model and crouch human body model.The fusion of parts of body detect result,we use the improve of the spring model. Because that the realation of position of parts of human conforms the Gaussian distribution we uses these three models to train a human relation ratio model.We describes these five parts'destribution as a human Gaussian distribution function through analyzing a large mount of sample data.In the process of detection, the result of a detection window can be computed out a ratio by these models.Then it can tells if there is people in the image by the threshold.The paper has relized all the methods above. We verify the system in the paper can detect human in the condition of standing,sitting,lying and dodging.It was tested on pedestrian data set,testing data set that was collected in simulated disaster site and real disaster site image set. The system is final embedded on the robotics platform of system of life rescue in disaster sit.The result shows that this method has reached a good effect.