Study On The Key Technologies Of Rapid Human Eye Detection

As one of the most important human sensory organs, the eye of human is the main information-gaining channel of human for perceiving the real world. Meanwhile, human eye is also the most salient feature of the human face which has an important function of information expressing and exchanging. The computer-vision-based human eye analysis is not only helpful for the researching of human visual system and understanding the complex psychological activities of people, but also an important step in human face analysis technology like face recognition and facial expression recognition. It has been widely used in human-computer interaction, driver assistance system, identity recognition, cognitive psychology and other fields. A lot of applications based on human face and eye analysis are on the condition that accurate and robust human eye detection is available, therefore, as the basic problem of human face and eye analysis human eye detection has great research significance.Currently, a lot of research results on human eye detection have been achieved. Firstly, this paper makes a comprehensive introduction of the foreign and domestic research status: a) Existing human eye detection methods are classified into several sorts and some representative methods of each category was introduced. b) Comparatively objective evaluation is made for some mainstream methods. c) For the shortcomings of existing methods, the development trend of the human eye detection in the feature is presented.Secondly, the popular Ada Boost-based rapid object detection framework is described in detail: a) The Haar-like feature and its fast computation using integral image is introduced. b) The development history of Ada Boost algorithm and its fundamental theory is presented. c) The weak leaner Stump and the structure of Cascade are described.Thirdly, an improved rapid eye detector based on Ada Boost object detection framework is proposed: a) a group of extended Haar-like features which can describe the human eye better are used to train the eye classifier. b) A method based on human eye region judgment and scale feature object searching strategy is proposed to accelerate the detecting speed. c) A hybrid method with face detection and eye detection is proposed to improve the detecting performance. At last an eye detection system is developed using C/C++ with Open CV library. The experimental result shows that the improved method have a much less false positives and a faster execution speed which means the system is fast and efficient in detecting and locating human eyes and especially useful in the real-time and resource-restricted applications.