基于自适应显著特征选择的动态加权平均行人识别模型

在跨场景行人识别过程中,为了解决多种特征以一个固定的权重融合导致行人识别率低、识别速度慢的问题,提出基于自适应特征选择的动态加权平均排名行人识别方法。首先,将GrabCut算法和基于流形排序显著性检测算法相融合,提高行人外观特征提取的准确性;然后,提出自适应显著特征选择方法,有效地提取行人特征描述;最后,通过动态加权平均排名模型将多特征融合。实验表明,所提出的方法提高了行人识别的准确性,同时对姿态的变化具有较好的鲁棒性。

A dynamic weighted average pedestrian identification modle based on adaptive feature selection

In order to solve the problem of low recognition rate and slow recognition speed in pedestrian identification, we propose a dynamic weighted average ranking pedestrian identification method based on adaptive feature selection to solve the problem of pedestrian recognition. Firstly, the GrabCut algorithm is combined with the manifold-based saliency detection algorithm to improve the accuracy of pedestrian appearance extraction. Then, we propose an adaptive selection method to effectively extract pedestrian features. Finally, we design a dynamic weighted average ranking model to merge multidimensional dynamic features. Experimental results show that the proposed method can improve the accuracy of pedestrian recognition, and has good robustness to the change of attitude.