Research On Human-hand Detection In Disaster Scene

Recent years, the rescue robot in the role of post-disaster relief work has been increasingly prominent, how to be better applied to the disaster relief efforts for survivors in many research institutions at home and abroad has been a common goal. Integrating the robot technology, disaster science, digital image science and other multi-disciplinary knowledge for researching and developing search and rescue disaster relief robot is a challenging new area of robotics research.Visible under the conditions of the manpower-based detection technology is part of the Vision Based Gesture Recognition technology, sign language reco- gnition and human-based target detection.In this paper, with the test of rescue robots in the images, we systematically study the visible part of the inspection manual under the conditions of disaster scene. This article focuses on the visible part of the hands test under the conditions of disaster scene, including: static images and moving images'detection. Firstly, based on every image testing method of hands under visible conditions, we should compare their application environment and advantages. To analyze the differences between a disaster scene images and daily life images, and then for some characteristics of hands, such as: the diversity attitudes, uncertainty and hands'color differences. Robots are easily influenced in complex background of interference factors and disaster, because hands are usually buried so that they can not render a complete shape. The study and propose effectively hands characteristics, then achieve the treatment of hands'image features. On one hand, this article uses color and texture features as the effective features of hands'static image, then decrypts them effectively and extracts by using the appropriate feature detecting method respectively. On the other hand, hands'detection in the video sequences also need to use the motion feature, this paper use space-time integration of the object segmentation algorithm and inter-frame difference time-domain segmentation algorithm to calculate changes in two adjacent frames. Airspace segmentation uses Canny segmentation edge detection method, after treatment by the morphological operators to form clumps in aggregate form. Based on priority fusion methods to combine the spatial-temporal information and then get exercise area. Using hand static image test methods to deal with motion region. Finally using SVM classifier.experimental extracted eigenvector of testing, about the disaster site occlusion detection rate reached more than 25% of the testing results of hands.