多特征融合与独立测度学习的行人再识别

发布时间：2018-06-22 03:06

本文选题：行人再识别 + 多特征融合　；参考：《中国图象图形学报》2016年11期

【摘要】：目的由于行人图像受到光照、视角、遮挡和行人姿态等变化的影响,在视觉上容易形成很大的外观差异,对行人再识别造成干扰。为了提高行人再识别的准确性,针对以上问题,提出一种基于多特征融合与独立测度学习的行人再识别算法。方法首先通过图像增强算法对原始图像进行处理,减少因光照变化产生的影响,然后对处理后的图像进行非均匀分割,同时提取行人图像的HSV、RGS、LAB和YCb Cr 4种颜色特征和SILTP(scale invariant local ternary pattern)纹理特征,在基于独立距离测度学习方法下,融合行人的多种特征,学习得到行人图像对的相似度度量函数,最后将行人图像对的相似度进行加权匹配,实现行人再识别。结果在VIPe R、i LIDS和CUHK01这3个数据集上进行实验,其中Rank1(排名第1的搜索结果即为待查询人的比率)分别达到42.7%、43.6%和43.7%,Rank5(排名前5的搜索结果中包含待查询人的比率)均超过70%,识别率有了显著提高,具有实际应用价值。结论提出的多特征融合与独立测度学习的行人再识别算法,能够有效表达行人图像信息,且对环境变化具有较强的鲁棒性,有效提高了识别率。
[Abstract]:Objective due to the influence of illumination, angle of view, occlusion and pedestrian posture on pedestrian images, it is easy to form great visual differences and interfere with the recognition of pedestrians. In order to improve the accuracy of pedestrian rerecognition, a pedestrian rerecognition algorithm based on multi-feature fusion and independent measure learning is proposed. Methods firstly, the original image was processed by image enhancement algorithm to reduce the influence of illumination change, and then the processed image was segmented inhomogeneously. At the same time, four color features and silTP (scale invariant local ternary pattern) texture features of HSVV-RGSlab and YCbCr are extracted from pedestrian images. Based on the learning method of independent distance measure, the multi-features of pedestrians are fused, and the similarity measure function of pedestrian image pairs is obtained. Finally, the similarity of pedestrian image pairs is weighted to achieve pedestrian recognition. Results the experiments were carried out on the three data sets of VIPe rimani lids and CUHK01. Among them, Rank1 (the ratio of the first search result to the person to be queried) reached 42.7% and 43.7%, respectively, and Rank5 (the ratio of people to be queried in the top 5 search results) exceeded 700.The recognition rate has been improved significantly and has practical application value. Conclusion the proposed multi-feature fusion and independent measure learning algorithm can effectively express pedestrian image information, and it is robust to environmental change and can effectively improve the recognition rate.
【作者单位】：合肥工业大学计算机与信息学院;
【基金】：国家自然科学基金项目(61371155) 安徽省科技攻关基金项目(1301b042023)~~
【分类号】：TP391.41

【相似文献】