Research On Aerial Target Tracking Based On Image

Aerial target tracking based on image regards image captured from air-borne or missile-borne CCD as its research content. The target will be segmented from the complex background image, which will use the image analysis algorithm. The data information gained from the algorithm will be used as guiding of air-borne or missile-borne weapons.This paper studied target tracking algorithm based on Gabor-Wavelet feature extraction and Gauss Mixture Model(GMM) classifier after the classification of current target tracking algorithms, the main work is as follows:(1)First, the common target tracking algorithm can be classified into four categories,Classification of currently most used target tracking algorithms is achieved, namely domain match, feature points match, active contour tracking, optical flow. Analysis of math-model construction, target pixels preprocessing and the performance of algorithm are accomplished to gain the features and weaknesses of each algorithm.(2)According to the advantages and disadvantages of the 4 types of algorithms discussed above, the paper designs the aerial target tracking algorithm based on Gabor wavelet and mixed Gauss model classifier. The captured image would be feature extracted by Gabor filter.This step divided into two categories that the generation of positive and negative training samples, testing samples and of positive, negative training samples and testing samples of the Gabor wavelet feature extraction; Gabor wavelet feature vector file extracted from the positive and negative samples trains Gaussian mixture classifier, which need set the maximum number of training and the training error requirements, further to determine the various Gaussian mixture model, weight coefficient and Gaussian distribution parameters;Then the image will be segmented by region grow algorithm. In the end, GMM will mark the target and gain its coordinate data.(3)Target tracking algorithm in this paper based on the Gabor wavelet feature vector extraction and hybrid Gauss model classifier, the accuracy and performance of which has been tested on the platform of Microsoft Visual Studio 2013 and OpenCV2.4.9, including float calculation model, function blocks decomposition, realization of C++ language, code implementation level and system level optimization to improve the efficiency of the algorithmand so on. The test result on X86 PC proved that the algorithm proposed in this paper has good accuracy and effectiveness.