Key Techniques On Infrared Small Target Detection And Track In The Complex Background

Infrared small target detection and tracking in the complex background has extensive application prospects in many fields of military and civilian situations. However, the infrared small target's pixel number and detail feature are very small, which makes it difficult to form an effective distinction between the infrared small target and the clutter. The lose track or tracking to the wrong objects are very common in the infrared small target detection and tracking system. These make the technology of infrared small target detection and tracking develop hardly, and the tracking distance of an optical tracking system is seriously affected. So this paper will focus on how to achieve stable and reliable tracking of infrared small target. This paper will strart from some key technical aspects, and solve these problems to achieve the stable and reliable detection and tracking.First, the preprocessing technique is studied. Here, the preprocessing technique concludes the nonuniformity correction and background suppression. The Space Low-pass and Temporal High-pass nonuniformity correction algorithm, the high-frequency Constant-Statistics Constraint nonuniformity correction algorithm, and the stripe nonuniformity correction algorithm based on the optimization techniques are presented in this paper. These scene based nonunformity correction algorithms make the convergence speed change from thousands of frames to tens of freames. To the stripe nonuniformity, it only needs one frame. In background suppression, the complex filter bank is introduced to the small target detection. It can increase the signal-to-clutter ratio.In this paper, the core structure of the Bayesian framework in the infrared small target tracking is constructed. The math model of the track point and observation probability is constructed. The multi-feature concept is introduced to this paper. The multi-feature observation probability and multi-feature track probability are designed. The introduction of the multi-feature enriches the trick of data association, and increases the efficiency of data association. Then the multi-filter technique is presented. Compared with the multi-model algorithm in mobile target tracking, this technique can save processing time and get good tracking effect.In this paper, five full small target tracking algorithms are presented, which includes the maximum observation probability algorithm, the multi-feature Probabilistic Data Association(PDA) algorithm, the multi-filter maximum observation probability algorithm, the particle Multiple-Hypothesis Tracking(MHT) algorithm, and the limited transfer probability particle filter algorithm. These algorithms are the realization of the Bayesian framework. The maximum observation probability algorithm, and the multi-feature PDA algorithm introduce the multi-feature to their original algorithms. These features are designed for small target detection, so they can increase the tracking effect. The multi-filter maximum observation probability algorithm is the combination of the multi-filter algorithm and the the maximum observation probability algorithm. It makes the maximum observation probability algorithm can work in mobile target tracking. The particle MHT algorithm abandons the back-search problem of the original MHT algorithm. And the MHT algorithm is changed to a Bayesian framework algorithm. This algorithm's efficeiency is enhanced and the processing time is saved. The limited transfer probability particle filter algorithm introudes the small target's high-frequency feature to the particle filter algorithm, and improves the design of the transfer probability. This improvemet can make a large number of particles no longer wast on useless area, and decrease the number of needed particles.This paper presents Image Processor Group Interface technique and the multi-processor multi-threading technique. The two hardware platform techniques solve the problem of multiple algorithms'parallel processing. It makes the small target tracking algorithm implementing in hardware platforms in real-time become possible.This paper starts from some key techniques of the of small target detection and tracking in complex background, and solves a variety of troubled problems, and promotes small target detection and tracking technology's development at last.