Research On Human Detection Based On Adaptive Features Selection

Human detection has become an important research in the pattern recognition field, it will be widely used in the future for the advantage of computer vision. For example, image retrieval, intelligent surveillance, digital entertainment, human computer interaction, sports analysis and other fields. The research on human detection based on adaptive features has important practical significance.For human movements have many complex actions, so they can be represented by many different features. It will not have perfect results if there is only a kind of feature is used to detect the moving human who are in different environments with different actions. For this problem, a new idea is held forth in this paper that is adaptive features selection. For this algorithm, firstly, the environment should be analyzed; secondly, the direction should be judged and lastly the information about the distance that the human had walked should be got by the system. Then the most suitable feature will be selected adaptively by the system. The experimental results indicate that this method is fast, dependable, and easily-operational.The Visual C++ IDE is used as the platform to create a system for human detection in this paper. It mainly depends on two features to do the human detection adaptively which are head-shoulder model and face’s feature. For building the head-shoulder model, firstly, the local maximum point and the object’s width of the binary image should be found in this algorithm by vertical projection histogram and horizontal projection histogram respectively. Through the local maximum point, the head-shoulder model’s height can be got. Then the model’s Hu Moment characteristic vectors can be extracted, and find out the similarity with the normative head-shoulder model which is got by a large number of tests. Finally, according to the threshold that set before the system will confirm if the object is a people. For the face detection, the cascade of classifiers from OpenCV is mainly used in this paper, this classifier is got by the algorithm of Adaboost.