Research On Infrared Target Detection And Tracking Under Complex Backgrounds

Homing guidance information processing system (HGIPS) is the hard core ofinfrared guided weapons, it takes on target detection and recognition in complexinfrared background. To improve the performance of infrared target detection andtracking in real HGIPS, this paper carries on researches on infrared dim point targetdetection, infrared area target detection and precise tracking of infrared target undercomplex background. The details are presented below.Stability of images in consecutive frames is required in temporal-filter-basedbackground suppression algorithms, while transformation-domain-based algorithmsbear heavy computation cost, and the spatial-filter-based algorithms can not predict thebackground accurately as well as enhance the target. To obtain a better suppressionperformance, a new complex background suppression algorithm based on fusion ofmorphological open filter and nucleus similar pixels bilateral filter is proposed in thispaper. The nucleus similarity degree is analyzed first, and then the two filters are fused.Experimental results demonstrate that the complex background suppression and targetenhancement can be accomplished more effectively, and the algorithm provides a goodbasis for target detection.Thresholding based segmentation of infrared images are widely used due to thecharacteristics of easy implementation, less calculation and high stability. However, theexisting thresholding based methods are performed depending on the assumption thatthe intensity difference between the background and the target is already known, so theability of bipolar targets segmentation in infrared image is restricted. To solve thisproblem, an adaptive segmentation algorithm for bipolar ship target detection in infraredimages is proposed. Multi-scale local standard-deviation-entropy variety is calculated toget rid of the assumption for intensity difference. Then a newmaximum-two-dimensional-entropy algorithm and fine segmentation algorithm areproposed to extract the ship targets. Experimental results demonstrate that the proposedalgorithm can get good performance in bipolar target segmentation.High false alarms probability goes with high detection probability. How to detectthe target trajectory and obtain real-time ability when false alarms probability is highare difficult for HGIPS. A trajectory detection algorithm which based on fast connectedcomponent features extraction and two-level trajectory association is proposed to solvethe problems. To extract the features of connected component more quickly, a hardwareacceleration based fast connected component labelling algorithm and a single pass fastconnected component analysis algorithm are proposed, the proposed algorithms aresuitable for hardware implementation (e.g. implement on FPGA), the experimentalresults demonstrate that the implementation in HGIPS can achieve real-time connected component labeling and connected component analysis. To improve the efficiency oftrajectory association algorithms under high false alarms probability condition, a targetdetection algorithm which based on two-level trajectory association is proposed. First,the short trajectories are detected in consecutive frames, and then a new trajectoryassociation which manage trajectories depends on the result of short trajectory detectionis applied. The analysis and experimental results demonstrate that the amount of invalidtrajectories can be decreased effectively while the targets are being detected accurately,which will leads to the efficiency improvement in target detection.Maneuver of target in a nearer distance and jitter of detector make the modeling oftarget trajectory very difficult, and the model-based-predict-and-association algorithmcan’t predict the trajectory effectively. A Sliding Window Kernel Ridge Regression(SWKRR) based trajectory predicting algorithm is proposed, and a better predictionperformance is achieved by transforming the two dimensional coordinates into a higherdimensional kernel space. The experimental results demonstrate that the proposedalgorithm can predict the nonlinear trajectories accurately, and the prediction error issmall. To reduce the influence of heavy cluttered background in mean-shift based targettracking, an Improved Multi-features Fusion based Mean-Shift (IMFMS) is proposed tolocate the target more precisely. Based on the tracking results of multi-features fusionbased mean-shift, iterative segmentation in local area is performed and the center ofregion is extracted, then the target can be located accurately. Based on the segmentationresult, the update scheme for template bandwidth and target model are applied.Therefore, the target can be tracked more accurately and robustly. By combine theSWKRR and IMFMS, a completed predict-and-matching tracking algorithm is obtained.Experimental results demonstrate that the proposed algorithm gets a stable performanceand higher efficiency.