Person Reidentification by Kernel PCA Based Appearance Learning
Jun Yang, Zhongke Shi, Patricio Antonio Vela · 2011
Person reidentification is a desirable feature in intelligent visual surveillance systems. This paper presents a novel person reidentification algorithm using an eigenspace appearance representation. Color and spatial information per each pixel form the pixel-wise appearance representation. Assuming a person's appearance under different poses, illumination condition and view points resides in a high dimensional, non-linear manifold, Kernel PCA is applied to represent the manifold. The similarity measurement is taken by projecting the testing data to the eigenspace representation. For efficient appearance representation, a key frame selection method is presented to select multiple representative templates for each person. Experimental results on two publicly available datasets demonstrate the performance of the proposed method.