| With the development of computer, network, communication and mobile Internettechnology, image understanding and computer vision technology are becoming the hot issuesof today’s research, such as moving targets identification of the military field, medical imagerecognition of the medical field, some industrial products identification in industrial areas,and so on. Target recognition which is one of the important tasks of image understanding andcomputer vision, is naturally becoming an important part of computer vision. The currentstudy is no longer still pictures, but the videos and a series of surveillance technology, such asintelligent monitoring systems, mobile robots, intelligent monitoring of traffic intersection.The author’s main job is based on pattern recognition and pattern classification, by reading therelevant literature, presents a research of target recognition in the video, the main workincludes the following three parts:First work in the paper, we are not concerned about all the goals for an input video usedto doing a simulation experiment, but the part which is the goal of the movement that hassignificant visual attention. Therefore, moving object detection and extraction is needed.Chapter2introduces the moving object extraction method. First, this chapter gives a briefoverview of the moving object extraction and introduces two relatively common methods ofmoving target detection, frames subtraction and background subtraction. By comparing theadvantage and disadvantage of both methods, author selects the background subtraction as therealization of this article. Secondly, the background model is the key to the implementation ofbackground subtraction, this paper analyzes the two background models of construction, aswell as the Gaussian distribution methods, and the author proposes a new background modelbased on frames subtraction which can build and automatically update the background.Finally, the motion detection needs pretreatment, dilation and erosion as two morphologicalmethods is used in this article.Second work in this paper, the author achieves the feature representation of the movingtarget. The author uses the method described by the localized features and uses Harris detectorto extract the feature points in the target, also uses the SIFT method descriptor indicates thedirection of distribution and the feature point, which reduces the time of calculating featurepoints, and the method also overcomes the light, rotation, translation to ensure the scaleinvariance feature representation and robustness. Finally, the moving target is represented bythe bag-of-words (BOW model). BOW model is represented as a document commonly used inthe field of information retrieval fields, now has a strong advantage in the field of imageunderstanding. BOW model mainly includes image feature extraction, feature representation,word list generation. We can complete the model representation of the moving target throughthe three steps. Third work in the paper, the author analyses and designs the classification model. Aswell known, a video sequence may contain many moving targets, for example, pedestrians,vehicles, and so on. We need do target classification to distinguish these different classes oftargets accurately. Chapter4of the paper analyzes and designs the classification model. First,the author selects the appropriate images as training samples, and founds the appropriateclassification model by training. In the paper, the author selects six image classes, faces,airplanes, car side, piano, butterfly and motorcycles. The paper selects AdaBoost classifier asmachine learning classification method, compares classification results and analyses andevaluates the classification performance, which provides a technical support for targetrecognition.At the end of the paper, the author does a simulation experiment for an input video, andanalyzes the application of this research method, which indicates the method can be applied inthe real intelligent monitoring system. Video processing independently is becoming animportant research subject while intelligent monitoring system is a major trend of thedevelopment in the future. |
