一種基于等距度量學習策略的行人重識別改進算法

周智恒; 劉楷怡; 黃俊楚; 陳增群

doi:10.11999/JEIT180336

一種基于等距度量學習策略的行人重識別改進算法

doi: 10.11999/JEIT180336

華南理工大學電子與信息學院 ??廣州 ??510000

基金項目: 國家自然科學基金(U1401252, 61871188)，國家重點研發(fā)計劃(2018YFC0309400)，中央高?；究蒲袠I(yè)務(wù)費專項資金(2017MS062)，廣州市產(chǎn)學研協(xié)同創(chuàng)新重大專項(201604016133)

詳細信息

作者簡介:
周智恒：男，1977年生，教授，博士生導師，研究方向為模式識別與人工智能

劉楷怡：女，1994年生，碩士生，研究方向為圖像處理與模式識別

黃俊楚：男，1994年生，博士生，研究方向為圖像處理與模式識別

陳增群：男，1995年生，本科生，研究方向為圖像處理與模式識別

通訊作者:
周智恒　zhouzh@scut.edu.cn

中圖分類號: TP391.41
計量
- 文章訪問數(shù): 2347
- HTML全文瀏覽量: 599
- PDF下載量: 93
- 被引次數(shù): 0
出版歷程
- 收稿日期: 2018-04-11
- 修回日期: 2018-09-13
- 網(wǎng)絡(luò)出版日期: 2018-09-20
- 刊出日期: 2019-02-01

Improved Metric Learning Algorithm for Person Re-identification Based on Equidistance

School of Electronic and Information Engineering, South China University of Technology, Guangzhou 510000, China

Funds: The National Natural Science Foundation of China (U1401252，61871188), The National Key R&D Program of China (2018YFC0309400), The Fundamental Research Funds for the Central Universities SCUT (2017MS062), Guangzhou City Science and Technology Research Projects (201604016133)

摘要

摘要:
為了提高行人重識別距離度量MLAPG算法的魯棒性，該文提出基于等距度量學習策略的行人重識別Equid-MLAPG算法。 MLAPG算法中正負樣本對在映射空間的分布不均衡導致間距超參數(shù)受負樣本對距離影響更大，因此該文設(shè)計的Equid-MLAPG算法要求正樣本對映射成為變換空間中的一個點，即正樣本對在變換空間中距離為零，使算法收斂時正負樣本對距離分布不存在交叉部分。實驗表明Equid-MLAPG算法能在常用的行人重識別數(shù)據(jù)集上取得良好的實驗效果，具有更好的識別率和廣泛的適用性。
- 行人重識別 /
- 等距度量 /
- MLAPG算法
Abstract:
In order to improve the robustness of MLAPG algorithm, a person re-identification algorithm, called Equid-MLAPG algorithm is proposed, which is based on the equidistance measurement learning strategy. Due to the imbalanced distribution of positive and negative sample pairs in the mapping space, sample spacing hyper-parameter of MLAPG algorithm is more affected by the distance of negative sample pairs. Therefore, Equid-MLAPG algorithm tends to map the positive sample pair to be a point in the transform space. That is, the distance of a positive sample pair in the transform space is mapped to be zero, resulting in no intersection in the distribution of positive and negative sample pairs in the transform space when algorithm convergences. Experiments show that the Equid-MLAPG algorithm can achieve better experimental results on commonly used person re-identification datasets with better recognition rate and wide applicability.
- Person re-identification /
- Equidistance /
- MLAPG algorithm

HTML全文

圖 1 MLAPG算法中$\mu $取值和訓練過程中所有樣本對馬氏距離均值對比示意圖

下載: 全尺寸圖片幻燈片

圖 2 對數(shù)邏輯損失函數(shù)變化趨勢

下載: 全尺寸圖片幻燈片

圖 3 在不同限制條件下正負樣本對距離情況

下載: 全尺寸圖片幻燈片

圖 4 正負樣本分布區(qū)域重疊示意圖

下載: 全尺寸圖片幻燈片

圖 5 交換空間中樣本分類情況

下載: 全尺寸圖片幻燈片

圖 6 VIPeR數(shù)據(jù)集上Equid-MLAPG算法與其他距離度量算法CMC曲線圖

下載: 全尺寸圖片幻燈片

圖 7 CUHK01數(shù)據(jù)集上Equid-MLAPG算法與其他距離度量算法CMC曲線圖

下載: 全尺寸圖片幻燈片

表 1 CUHK03數(shù)據(jù)集上多種距離度量算法對比

算法	檢測標注			人工標注
算法	第1匹配率(%)	第5匹配率(%)	第10匹配率(%)	第1匹配率(%)	第5匹配率(%)	第10匹配率(%)
XQDA	46.25	78.90	88.55	52.20	82.23	92.14
MLAPG	51.15	83.55	92.05	57.96	87.09	94.74
Nullspace	53.70	83.05	90.30	58.90	85.60	92.45
Equid-MLAPG	52.41	85.25	92.84	58.72	89.07	95.28

下載: 導出CSV

表 2 Marlet1501，DukeMTMC-reID數(shù)據(jù)集上多種距離度量算法對比

算法	Market1501數(shù)據(jù)集		DukeMTMC-reID數(shù)據(jù)集
算法	第1匹配率(%)	平均準確率(%)	第1匹配率(%)	平均準確率(%)
XQDA	43.23	22.00	31.37	17.17
MLAPG	42.52	21.45	36.58	19.10
Nullspace	54.60	29.80	45.02	26.11
Equid-MLAPG	44.25	24.38	39.25	21.54

下載: 導出CSV

參考文獻(22)

ZHENG Liang, YANG Yi, and HAUPTMANN A G. Person re-identification: Past, present and future[OL]. arXiv preprint arXiv: 1610.02984, 2016.

SHAH J H, LIN Mingqiang, and CHEN Zonghai. Multi-camera handoff for person re-identification[J]. Neurocomputing, 2016, 191: 238–248. doi: 10.1016/j.neucom.2016.01.037

REHMAN S U, CHEN Zonghai, RAZA M, et al. Person re-identification post-rank optimization via hypergraph-based learning[J]. Neurocomputing, 2018, 287: 143–153. doi: 10.1016/j.neucom.2018.01.086

PEDAGADI S, ORWELL J, VELASTIN S, et al. Local fisher discriminant analysis for pedestrian re-identification[C]. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Protland, USA, 2013: 3318–3325.

WEINBERGER K Q, BLITZER J, and SAUL L K. Distance metric learning for large margin nearest neighbor classification[C]. Advances in Neural Information Processing Systems. Vancouver, Canada, 2006: 1473–1480.

DAVIS J V, KULIS B, JAIN P, et al. Information-theoretic metric learning[C]. Proceedings of the 24th International Conference on Machine Learning, Corvalis, USA, 2007: 209–216.

DIKMEN M, AKBAS E, HUANG T S, et al. Pedestrian recognition with a learned metric[C]. Asian Conference on Computer Vision, Queenstown, New Zealand, 2010: 501–512.

ZHENG Weishi, GONG Shaogang, and XIANG Tao. Person re-identification by probabilistic relative distance comparison[C]. Computer Vision and Pattern Recognition, IEEE, Colorado, USA, 2011: 649–656.

KOESTINGER M, HIRZER M, WOHLHART P, et al. Large scale metric learning from equivalence constraints[C]. Computer Vision and Pattern Recognition (CVPR), Rhode Island, USA, 2012: 2288–2295.

TAO Dapeng, JIN Lianwen, WANG Yongfei, et al. Person re-identification by regularized smoothing kiss metric learning[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2013, 23(10): 1675–1685. doi: 10.1109/tcsvt.2013.2255413

LIAO Shengcai, and LI S Z. Efficient PSD constrained asymmetric metric learning for person re-identification[C]. Proceedings of the IEEE International Conference on Computer Vision. Santiago, USA, 2015: 3685–3693.

NESTEROV Y. Introductory Lectures on Convex Optimization: A Basic Course[M]. New York, USA, Springer Science & Business Media, 2013: 15–20.

TSENG P. On accelerated proximal gradient methods for convex-concave optimization[OL]. http://www.mit.edu/~dimitrib/PTseng/papers/apgm.pdf.

GRAY D and TAO Hai. Viewpoint invariant pedestrian recognition with an ensemble of localized features[C]. European Conference on Computer Vision, Marseille, France, 2008: 262–275.

LI Wei, ZHAO Rui, and WANG Xiaogang. Human reidentification with transferred metric learning[C]. Asian Conference on Computer Vision. Daejeon, Korea, 2012: 31–44.

LI Wei, ZHAO Rui, XIAO Tong, et al. Deepreid: Deep filter pairing neural network for person re-identification[C]. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Columbus, USA, 2014: 152–159.

ZHENG Liang, SHEN Liyue, TIAN Lu, et al. Scalable person re-identification: A benchmark[C]. Proceedings of the IEEE International Conference on Computer Vision. Santiago, USA, 2015: 1116–1124.

ZHENG Zhedong, ZHENG Liang, and YANG Yi. Unlabeled samples generated by GAN improve the person re-identification baseline in vitro[C]. IEEE International Conference on Computer Vision. Venice, Italy, 2017: 3774–3782.

LIAO Shengcai, HU Yang, ZHU Xiangyu, et al. Person re-identification by local maximal occurrence representation and metric learning[C]. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Boston, USA, 2015: 2197–2206.

ZHANG Li, XIANG Tao, and GONG Shaogong. Learning a discriminative null space for person re-identification[C]. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Las Vegas, USA, 2016: 1239–1248.

曾明勇, 吳澤明, 田暢, 等. 基于外觀統(tǒng)計特征融合的人體目標再識別[J]. 電子與信息學報, 2014, 36(8): 1844–1851. doi: 10.3724/SP.J.1146.2013.01389

ZENG Mingyong, WU Zeming, TIAN Chang, et al. Fusing appearance statistical features for person re-identification[J]. Journal of Electronics &Information Technology, 2014, 36(8): 1844–1851. doi: 10.3724/SP.J.1146.2013.01389

MATSUKAWA T, OKABE T, SUZUKI E, et al. Hierarchical gaussian descriptor for person re-identification[C]. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas, USA, 2016: 1363–1372.

相關(guān)文章

施引文獻

資源附件(0)

訪問統(tǒng)計