Dim Moving Target Tracking Algorithm Research Based On Sparse Representation

The infrared dim target detection and tracking technology is widely used in areas of infrared missile and warning, video surveillance, medical detection and other fields. It owns important military and social value to research a new method on tracking infrared dim target under the condition of low SNR. In recent years, the theory of image sparse representation theory attracts more and more attention because of its simplicity and efficiency in signal representation. It adopts redundant over-complete dictionary in signal representation instead of the traditional Fourier transform, Wavelet transform based on the mathematical basis functions. The over-complete dictionary finds the least and optimal atom constructed the signal which is good for target recognition.This thesis has further studied on the infrared dim target tracking technology based on signal sparse representation theory, the main research results are:① An adaptive morphological ingredients over-complete dictionary based on content-learning approach is proposed to building a dictionary which can represent accurately small dim target morphology. The third chapter also compares the sparse representation capabilities of Gabor over-complete dictionary, Gaussian dictionary, and adaptive morphological ingredients dictionary on reconstructing the small dim target signal. Experimental results show that the adaptive morphological ingredients dictionary contains more various morphological ingredients of target atoms, which is more conducive to the reconstruction of dim target signal.② A small dim target detection algorithm based on adaptive online discriminative dictionary is proposed. This thesis discovers the sparse representation differences between target signal and background noise signal on Gaussian clutter over-complete dictionary, thus the adaptive morphological over-complete dictionary is further divided into the target over-complete dictionary and background over-complete dictionary by the Gaussian over-complete dictionary, which can enhance different features between target and background. The reconstructed residual energy differences that the signal represents in the target dictionary and background dictionary are used to detect small dim target signal, therefore it could improve the accuracy of small dim target detection.③ A small dim target tracking algorithm based on the particle filter framework was presented. The algorithm establishes observation model of small dim target based on online discriminative dictionary. The sparse reconstruction error differences between particle target image block and the particle background image block are used to estimate target state, and to maintain the stability of the target tracking. Moreover, this paper also presents a stochastic estimation method online updates dictionary subspace, which could improve the computational efficiency, real-time and robustness of tracking algorithms.