Research On Infrared Small Target Detection And Tracking Algorithm

After decades of research and exploration, the infrared small target detection and tracking is still a hot issue in infrared field.With the development of infrared technology and the higher request for air defense weapon system, the increased indicators such as requirement for target detection range and accuracy cause the size reduction of small target and more serious influence on detection and tracking in clutter environment, which requires deep research for small target detection and tracking algorithms. For these reasons, it is more urgent and meaningful to explore a class of algorithms that can deal with complex environmental changes, detect and track manoeuvrable target effectively and steadily under low SNR environment. Therefore, this article focuses on the infrared image pre-processing technology, the detection and tracking algorithms for manoeuvrable target, and the main research contents and innovative work are summarized as follows:Firstly, the infrared image pre-processing technology has been studied systematically,some kinds of common pretreatment methods are introduced, and the performance of several filtering algorithms is compared, among which the median filter and Top-hat filter have optimal effects. This paper improves the traditional Top-hat filtering algorithm and puts forward the structural elements center weighted Top-hat filter algorithm, which enhances the effect of filter greatly, then studies the influence of center weighted value on filtering effect.The experiments show that, at the beginning,the image SNR has a upward trend as the weighted value increases, but when n is too large, the appearance of target information will be destroyed, even worse, the target will be filtered out together with the background.Secondly, ameliorates the original adaptive threshold method based on the maximum search method in order to have a better segmentation effect. Through giving the overall maximum a certain weight, the influence on final threshold is increased, which improves the segmentation threshold, reduces the number of candidate targets in segmentation results, and cases the burden of small target detection and tracking tasks. After segmentation, the influence on target size can be recovered by area compensation operation.Thirdly, this paper researches and proposes an infrared small target detection algorithm that is the combination of morphological filtering and kinematic likelihood model. The improved Top-hat filtering algorithm and adaptive threshold segmentation are used to extract the candidate target. The motion model and motion likelihood model based on Kalman filtering incentive mechanism are applied, and this algorithm detects small target according to the movement likelihood of candidate target in two frames. Experimental results signify that this algorithm can achieve small target detection task under low SNR and complicated background.Finally, this paper proposes two small infrared target tracking algorithms. The first one is based on the Kalman filter and combines motion likelihood model with appearance likelihood model, establishes the small acceleration speed motion model based on Kalman filtering incentive mechanism, estimates trajectory of small target, and achieves the matching and tracking of small target through the joint probability of motion likelihood and appearance likelihood within the neighborhood of estimated position. Experimental results indicate that the algorithms can track maneuvering small target under low SNR and complicated background precisely and steadily, and adjust back to right track timely when target moves.The sencond target tracking algorithm is the infrared small target tracking algorithm based on the theory of sparse image. Apply the image sparse theory to obtain sparse coefficients of target image blocks and background image blocks based on the best representation of the over-complete learned dictionary. The sparse coefficients of target image blocks are significantly different from background image blocks so that we can obtain the trajectory of small target by building a new parametric model which correlates the probability of each candidate target will be the target to be tracked with sparse coefficients. Experiments show that the algorithm is suitable in a relatively high SNR environment of small target tracking problem under which we can achieve accurate tracking.