Reseach On Visual Object Recognition In Dynamic Scences

Object recognition is playing an increasingly important role both in military applications, home life, health, traffic control and other aspects of artificial intelligence. Visual object recognition is an irreplaceable aspect of machine vision, and it is an important way for a computer to have awareness. In recent years, the rapid development of object recognition technology is gradually shifting from the research stage to reality applications.This paper is a research of visual object recognition in dynamic scenes. According to the object in dynamic scenes easily influenced by illumination, perspective and background, we did researches on object detection and description, object recognition and etc. The main contents of this page are as follows:(1) We studied the research status and development trend of object recognition in our country and abroad,and then introduced some of commonly used object recognition model. Combined with the characteristics of the visual object, we choose the Bag of Feature model as a framework for our researches after analysis and comparisons.(2) In the module of object detection and description, based on the detailed analysis and comparison of local features-MSER and DOG, we choose the local features DOG to extract the key points.And then we adopted SIFT and BRIEF separately to implement description of the keypoints. By this, we build the features with characteristics of scale and rotation invariance to avoid that the object in dynamic scenes were changed easily in the aspect of geometric.(3) In the module of object recognition, we analyzed the support vector machine classifier based on histogram intersection kernel in detail. And then we compared it with the k-nearest neighbor classifier and the support vector machine classifier based on the RBF kernel. At last the feasibility of the support vector machine classifier based on histogram intersection kernel were verified by detailed experiments and analysis.(4) The Spatial Pyramid BOF model(Spatial Pyramid BOF, referred SPBOF) based on the BOF model were used in this paper. And then we conducted a lot of experiments compared with the BOF model and verified the feasibility of Spatial Pyramid BOF.(5) We proposed the algorithm of Spatial Pyramid BOF with SVM based on histogram intersection kernel. According to the experimental results compared with the traditional algorithms, the algorithm of this paper can achieve a good recognition rate in object recognition in dynamic scenes.