目标再确认中的优化扩散距离相似性度量

Target re-identification based on optimized diffusion distance

跨摄像头目标再确认是多摄像头监控领域中一个亟需解决的难点问题,如何获得准确率更高的距离度量算法成为解决该难点的关键。为此本文在提取优秀的多特征基础上,建立了一种无需训练,适应更多场景的度量算法:优化扩散距离相似性度量,用于跨摄像头目标再确认。在高维扩散距离的基础上,加入高斯金字塔图像采样和加权性优化处理,用于提高相似空间向量的辨别力,并提高计算效率。通过对高维扩散距离的二次优化建立起最终的相似性度量函数。最后对VIPe R和ETHZ数据库中的图片进行多次目标再确认实验,排名第一的图片的正确匹配率达到了50.5%。实验结果表明本文算法取得了较好的匹配结果。

Target re-identification via cross-camera is a challenging problem in the field of multi-camera surveillance.How to get a more accurate distance measurement algorithm is the key to solve this difficult problem.So,in this paper,we established a new measurement algorithm without training based on extracting more excellent features to do target re-identification,which is optimized diffusion distance.On the basis of high-dimensional diffusion distance,Gaussian pyramid image sampling and weight optimization are added to improve the discrimination of similar space vectors and increase computational efficiency.The final similarity measure function is established by the second optimization of the high-dimensional diffusion distance.At last,we do numerous target re-identification experiments based on databases VIPeR and ETHZ.The matching rate of rank first image can reach50.5%.The experimental results show that the algorithm proposed in this paper has good performance.