Real Time Hand Detection System Based On Co-occurrence Histogram Of Oriented Gradients

Hand detection is a hot but very difficult topic in the research of pattern recognition. Firstly, it has extremely wide applications, such as the Human-Computer Interaction, intelligent home alliance control; furthermore, it can even achieve person-to-person communication. Secondly, hand detection is still a hard topic though, as the hand is different from some other objects which are rigid and regular. On the contrary, the shape of a hand has great difference in various view and posture. In the worst situation, it can show many kinds of shape, which is very difficult for detection. So there is still no such an elegant approach in this field, and it is a task full of challenge and chance.The Histogram of Oriented Gradients (HOG) based method is getting more and more attention in the object detection field. The advantage of this feature is that it has good robustness under slight rotation and deformation. And as it uses the Support Vector Machine as the classifier, it has excellent ability of generalization and achieves good performance in some object detection approach especially the application of human detection. Based on the good performance of HOG, this paper adopts it as the feature for the hand detection system. Meanwhile, as it has been said in many others papers, the HOG feature still has many disadvantages, such as the speed is slow and the ability of generalization is weak. So in order to solve these problems, this thesis proposes some improvement. First, according to the characteristic of hands, we use more simplified models for detection, and adopt the method of multi-scale detection and tracking. Besides, this thesis proposes a new feature based on the feature of Co-occurrence Histogram of Oriented Gradients(CoHOG) to improve the accuracy of the system. And we use multi-model in the detection which can detect hands in various orientation and gesture. At last, an infrared camera is used to overcome effect of illumination, which improves the performance further.