基于改进SIFT特征和粒子滤波的目标识别仿真研究

摘要

针对在目标识别中原始SIFT(尺度不变特征转换)特征算法计算量大,特征点匹配耗时长等缺陷,采用一种改进的SIFT特征算法.在原始的SIFT算法基础上简化了特征描述符,以及对匹配算法进行了改进,考虑到识别过程中目标物体的特征点会发生变化,因此结合粒子滤波来实现对目标物体的识别.仿真结果表明:该方法继承了原始SIFT算法的优点,有效地避免了一些干扰,减小了计算量,在结合粒子滤波算法后能够有效地更新特征点的匹配,最终实现了对目标物体准确的识别.%For the problems of much calculation and consuming more time in the original SIFT feature algorithm, a kind of improved SIFT feature algorithm is used. Based on the original SIFT feature algorithm, the feature descriptor is simplified and the matching algorithm is improved. Considering the changing of feature descriptors in object recognition, the particle filter algorithm is combined with the target object recognition. Simulation result shows that this algorithm retains the advantages of original SIFT features algorithm, some disturbances are avoided and calculated amount is decreased . In combination with the particle filter algorithm it can effectively update feature point matching. The objects can be recognized reliably by using the combination algorithm.