中文摘要：

针对接力跟踪中目标跨摄像机识别困难的问题,提出了在傅里叶频域进行目标再识别的方法。通过对频谱的局部平均化处理,使非刚性目标的频谱在幅值上近似为其各刚性子部分频谱的简单求和。再通过最窄目标取样窗口限制各子部分的变化,使整体频谱大致不变。最后根据频谱各点幅值变化倍率的一致性判断目标的相似性,使频谱信息得到充分使用。实验表明,该方法的目标识别能力和鲁棒性相对于几种常见的基于纹理的方法有一定的提高。

英文摘要：

For the difficulty of identifying relay-tracked targets across cameras, a method for re-identifying targets in frequency domain is proposed. By locally averaging, the non-rigid target＇s Fourier spectrum is approximated by the real sum of its relatively rigid components＇ spectrum in amplitude. Then, by using the narrowest sampling window to limit the components＇ changes, the target spectrum＇s change is limited to a low degree. Finally, the consistency of the amplitude changing rate at every point of the spectrum map is used as the similarity metrics of targets, thus making full use of spectrum information. Experiments show an improvement on target recognition and robustness relative to the existing methods based on texture.