基于局部亮度直方图特征的实时目标识别与跟踪  

Real-time target recognition and tracking using local intensity histogram

作　　者：孙抗[1] 汪渤[1] 郑智辉[1] 

机构地区：[1]北京理工大学自动化学院,北京100081

出　　处：《系统工程与电子技术》2011年第9期1927-1931,共5页Systems Engineering and Electronics

基　　金：国家"十一五"预研项目(51309020402)资助课题

摘　　要：针对自寻的电视导引头实时跟踪需求及其硬件平台的限制,借鉴离线训练的思想,提出一种基于局部亮度直方图的二进制特征描述算子,利用该算子设计了基于差异度函数和索引值的两级特征匹配方法。算子的二进制特性使得特征匹配过程主要通过一系列二进制位运算完成,算法复杂度大大降低。实验表明,该算法实时性强,平均跟踪速度可达25 fps以上,运行过程中内存需求量小,并且对目标尺度变化、旋转、光照变化等具有较好的鲁棒性,满足了导引头实时目标识别与跟踪需求。To deal with the conflict between real-time target tracking and low-power hardware platform of television homing seeker, a binary feature descriptor using local intensity histogram is proposed, which is inspired by the off line training method. Then, a two-stage feature matching scheme consisting of dissimilarity function and index value is adopted based on the proposed descriptor. The main advantage of this scheme is to convert time consuming feature matching to simple bitwise operations. Experiment results demonstrate that the proposed method achieves outstanding runtime performance at the speed of over 25 fps with only a little memory cost. With its robustness to scale changes, rotation and illumination changes, the method could well fulfill the real-time target recognition and tracking needs of homing seekers.

关 键 词：电视导引头 特征匹配 目标跟踪 亮度直方图 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]