The Research And Application About The Detection Of Target On Variable Features In Computer Vision

Computer vision is used to imitate human vision for understanding the imagie what the computer has seen, and triggering some behaviors by the computer and auxiliary equipment. As an important research field of the computer science, Computer vision is wide and involves computer graphics, artificial intelligence, signal processing, etc. it also has a wide application prospect, such as the entrance guard system of face recognition, the industrial product quality inspection system, etc. Target detection is a direction of computer vision research, and has a decisive influence on the subsequent operation. There is a large number of domestic and foreign scholars to study target detection, and achieve rich results. On the basis of existing achievements, the paper focuses on the Haar features detection’s research and application what are about that the variable target to be detected by features under arbitrary movement.The paper instroduces the basic theory of Haar features, and does experiments about the detection of variable target when target and observation point and background all under arbitrary movement.Through the research and analysis of the Haar features detection, and find it has a problem. The problem is that the Haar features is a good method of detection, but when the target is variable it will can not find out the target as before. So the paper show a improved method about the Haar features detection to solve the problem. The traditional method of Haar features detection what be designed to learn of the existent samples’ features, but not to find out the regularity about the variation of samples. In the improved method, forecast as a new part be joined into the training of the Haar features cascade classifier. In the result, the classifier can detect and recognize the future features of the target at present. In addition, a new thing named Double-Cascadeclassifier what is used to control target in a reasonable range, and takes full adventage of the historical data’s value of convenience and proofreading.In the practical application, based on OpenCV, the system be combined with a model car. It achieves that the model car can be controled to track the target correctly by the detection system.