Study On Target Recognition And Tracking Based On Image

With the rapid development of computer vision technology and image parallel processing technology, visual target recognition and tracking technology is widely used in the military field, aerospace, scientific exploration, astronomical observation and civilian areas. Target recognition and tracking technology has become a hot topic of research in the automatic control field, computer vision field and pattern recognition field. It presents mainly on target recognition and tracking technology with practical research background.Firstly, it gives an adaptive fuzzy image filter algorithm in view of Gaussian and impulse noise. The algorithm demarcates impulse noise to separate the two kinds of noise in each part by noise detection. Then filter algorithm is applied to remove the impulse noise by using adaptive median filter window. Based on filtering the impulse noise, combining histogram and membership function in fuzzy control, an improved mean algorithm is given to filter gauss noise. It can overcome the faintness with effectively filter the gauss noise.Secondly, it presents the Gabor wavelet character recognition algorithm based on character space. It gives the new way to solve the key problem of Gabor wavelet based on analysis Gabor wavelet algorithm's characteristics and parameters. It gives a new algorithm to calculate Gabor wavelet coefficients and a new way to select Gabor wavelet's parameters. Aiming at the problem of Gabor wavelet's calculation large quantity, it is found the relationship between Gabor wavelet transform and discrete Hartley transform to realize the fast calculation algorithm. It has been significantly reduced in Gabor wavelet coefficients calculated volume. It is given a new Gabor wavelet parameter selection algorithm in this paper. It gives a new way to select Gabor wavelet's parameters by importing energy function and optimal function to the classic algorithm. It not only can exert the characters of Gabor wavelet's parameters, but also can avoid the calculation large quantity problem of the classic algorithm. Image matching feature points in the research of target matching, it gives a new concept of "Sub target" and "Sub template" . It changes the pattern of traditional image target matching tracking algorithm. Though it can not avoid the problem of image ergodicity, it can decrease the image matching calculation and reduce the computation time of the matching algorithm.The target recognition algorithm based on template matching, it presents an improved sequence similarity detection algorithm and an improved mean shift algorithm. The improved sequence similarity detection algorithm presents four recommendations and template updated strategy adaptively to enhance rapid and accuracy. Mean shift algorithm is a relatively new algorithm for target recognition with well real-time, positioning accuracy and robustness, although there are also as insufficiency for recognition of rapid movement, suffering large proportion block and the lack of template updated strategy. It gives the corresponding way to improve the insufficiency of the mean shift algorithm.Finally, it is presented a new nonlinear and adaptive filter algorithm based on the current model Kalman filter algorithm. Aiming at the insufficiency of current model Kalman filter algorithm with mobile target, it can effectively compensate the error of linear model, regulate the system gain in real time, improve the target tracking character and improve accuracy.