Aimed at the characteristics of infrared images, a new method was proposed for real-time infrared target tracking. The multi-scale Local Binary Pattern (LBP) histograms extracted by LBP operator was used as image feature vector, and the similarity value between template and sample image was calculated out by Chi-square test. By using Kalman filter together with multi-scale template updating mechanism, the method could keep steady tracking even when the target was varied widely in size, partly occluded or temporarily lost. The image processing board was designed, and hardware implementation was introduced, which satisfied the real-time target tracking requirement of 50 Hz/s. Experiment result showed the efficiency and robustness of the proposed algorithm.%针对红外图像的特点,提出一种实时目标跟踪方法.该方法采用局部二元模式(LBP)算子提取图像的多尺度LBP编码直方图作为图像特征向量,使用卡方统计度量模板图像和样本图像的相似度.结合Kalman滤波和多尺度模板更新机制,使得算法在目标尺寸大幅变化,目标部分遮挡、短暂消失等情况下仍能保持跟踪的正常稳定进行.针对该算法,设计了专用图像处理板卡,介绍该算法在图像处理板上的硬件实现方法,使目标跟踪满足50 Hz/s的实时处理要求.最后,使用两组实录红外序列图像进行了仿真实验,验证了该算法的有效性和鲁棒性.
展开▼
