Feature Fusion and Ellipse Segmentation for Person Re-identification
Person re-identification refers a task of associating the same person in different camera views.Due to the variance of camera angles,pedestrian posture and lighting conditions,the appearance of the same pedestrian in different surveillance videos might change greatly,which becomes a major challenge for person re-identification.To solve the above problems,this paper proposes a feature fusion and ellipse segmentation algorithm for person re-identification.First of all,in order to reduce the impact of changes in light illumination,an image enhancement algorithm is used to process pedestrian images.Then the ellipse segmentation algorithm is applied to reduce the influence of background clutter in the image.After that,we extract features which contain more abundant information and merge them together.Finally,bilinear similarity metric is combined with Mahalanobis distance as a distance metric function,and the final metric matrix is obtained by using optimization algorithm.Experiments are performed on three public benchmark datasets including VIPeR,PRID450s,CUHK01,and the results clearly show the significant and consistent improvements over the state-of-the-art methods.
Person re-identification Feature fusion Ellipse segmentation
Meibin Qi Junxian Zeng Jianguo Jiang Cuiqun Chen
School of Computer and Information,Hefei University of Technology,Hefei 230009,Anhui,China
国际会议
广州
英文
50-61
2018-11-23（万方平台首次上网日期，不代表论文的发表时间）