Research On Human Head Detection In The Disaster Site

The evolutionary history of human civilization is a history of constant struggle with the disaster. In recent years, many natural disasters have occurred, and the primary task after the disasters is to search and rescue the survivors. With the advances of modern science and technology, the robot is gradually applied in post-disaster search and rescue. The human head detection in the disaster site as the robots eyes' supporting technology, its research is very critical.Although the representative detection methods such as the Pedestrian detection methods, the face detection methods and the number plate detection methods and so on, are quite sophisticated and have acquired good results under normal circumstances. But the disaster site is quite different, in which objects disperse here and there, the background is complicated, the illumination conditions are complicated, and the human body's shading and distortion are serious. So the human head detection using only one feature and one method can't meet the requirements.This paper proposes and implements a new human head detection method based on the registration of visual and infrared images, and the combination of several features. These features include the skin color feature in the visual images, the brightness feature in the infrared images and the multi-scale orientation feature. Initially, the image registration method based on the corner feature is used to achieve the registration of the color image and the infrared image; secondly, the strong regions of interest and the weak regions of interest are extracted according to the skin color feature of the visual images and the brightness feature of the infrared images; and then, the multi-perspective classifiers are trained using the Adaboost algorithm based on the multi-scale orientation feature. Combination of image rotation, the multi-profile and the multi-view human head detection can be achieved; At last, the detection results are integrated. According to the test results on the testing sets and our own simulation disaster site sets, and comparing with other algorithms, it can proves that this method has got the more quick computing speed and the more accuracy of detection. This method can satisfy the need of the human head detection on the disaster site.