基于二进制稳健独立基元特征的扩展目标快速检测  

Fast Detection of Expanded Target Based on Binary Robust Independent Elementary Features

作　　者：邓黎[1,2] 王春鸿[1] 饶长辉[1] 

机构地区：[1]中国科学院光电技术研究所自适应光学重点实验室,四川成都610209 [2]中国科学院研究生院,北京100049

出　　处：《中国激光》2012年第B06期327-331,共5页Chinese Journal of Lasers

摘　　要：在某些扩展目标光电成像中，目标图像缺少局部细节，因此采用复杂的特征检测算法和高维特征描述符，但这种方法不仅存在特征描述区分度弱的问题，而且还存在资源占用多、运算速度慢以及难以实现实时处理的缺点。主解决此问题提出了用加速分段测试提取特征（FAsT）检测算法进行角点检测，用二进制稳健独立基元特征（BRIEF）描述符进行目标特征描述的新方法。同时，针对BRIEF描述符缺少方向判别，对目标姿态变化敏感的问题，提出了主方向约束机制，有效地提高了特征点识别的稳定性。将本方法与加速稳健性特征（suRF）和尺度恒定特征变化（SIFT）两种应用广泛的算法进行了比较，结果表明，本方法的运算速度分别达到了SURF的5倍和SIFT的17倍，且识别率与SURF相当，能在不降低特征识别率的基础上，实现目标的快速检测和稳定跟踪。In some opto-electrical imaging system for extended target, because the target is lack of local detail, using complex detector and high-dimensional descriptor not only brings the problem of weak discrimination of description, but also results in high sources occupation and slow calculation, lead to a hard process in real-time. A specialized algorithm is proposed with features from accelerated segment test （FAST） as the detector and binary robust independent elementary feature （BRIEF） as the descriptor, to solve the problems efficiently. In the meantime, because the BRIEF is sensitive to the changes of target, a process named major orientation constraint is employed which can improve the stability of feature recognition. Experiments are done and this new algorithm is compared to two widely used algorithms, speed-up robust feature （SURF） and scale invariant feature transform （SIFT）. The result shows that our algorithm performs 5 times of SURF and 17 times of SIFT in speed, meanwhile the recognition accuracy is comparable to that of SURF. Hence this algorithm achieves fast object detection and stable tracking without decline of recognition accuracy.

关 键 词：图像处理 目标检测 跟踪 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]