基于新型三元卷積神經(jīng)網(wǎng)絡(luò)的行人再辨識(shí)算法
doi: 10.11999/JEIT170803 cstr: 32379.14.JEIT170803
-
1.
(華僑大學(xué)工學(xué)院 泉州 362021)
-
2.
(華僑大學(xué)信息科學(xué)與工程學(xué)院 廈門 361021)
基金項(xiàng)目:
國(guó)家自然科學(xué)基金(61602191, 61401167, 61473291, 61605048, 61372107)，福建省自然科學(xué)基金(2016J01308)，廈門市科技計(jì)劃項(xiàng)目(3502Z20173045)，華僑大學(xué)中青年教師科技創(chuàng)新資助計(jì)劃(ZQN-PY418, ZQN-YX403, ZQN-PY518)，華僑大學(xué)科研基金資助項(xiàng)目(16BS108)
Person Re-identification Based on Novel Triplet Convolutional Neural Network
-
1.
(College of Engineering, Huaqiao University, Quanzhou 362021, China)
-
2.
(School of Information Science and Engineering, Huaqiao University, Xiamen 361021, China)
-
3.
(Institute of Automation, Chinese Academy of Sciences, Beijing 100190, China)
Funds:
The National Natural Science Foundation of China (61602191, 61401167, 61473291, 61605048, 61372107), The Natural Science Foundation of Fujian Province (2016J01308), The Scientific and Technology Funds of Xiamen (3502Z20173045), The Promotion Program for Young and Middle Aged Teacher in Science and Technology Research of Huaqiao University (ZQN-PY418, ZQN-YX403, ZQN-PY518), The Scientific Research Funds of Huaqiao University (16BS108)
-
摘要: 基于三元卷積神經(jīng)網(wǎng)絡(luò)的行人再辨識(shí)算法多數(shù)采用歐式距離度量行人之間的相似度，并配合鉸鏈(hinge)損失函數(shù)進(jìn)行卷積神經(jīng)網(wǎng)絡(luò)的訓(xùn)練。然而，這種作法存在兩個(gè)不足：歐式距離作為行人相似度，鑒別力不夠強(qiáng)；鉸鏈損失函數(shù)的間隔(Margin)參數(shù)設(shè)定依賴于人工預(yù)先設(shè)定且在訓(xùn)練過程中無法自適應(yīng)調(diào)整。為此，針對(duì)上述兩個(gè)不足進(jìn)行改進(jìn)，該文提出一種基于新型三元卷積神經(jīng)網(wǎng)絡(luò)的行人再辨識(shí)算法，以提高行人再辨識(shí)的準(zhǔn)確率。首先，提出一種歸一化混合度量函數(shù)取代傳統(tǒng)的度量方法進(jìn)行行人相似度計(jì)算，提高了行人相似度度量的鑒別力；其次，提出采用Log-logistic函數(shù)代替鉸鏈函數(shù)，無需人工設(shè)定間隔參數(shù)，改進(jìn)了特征與度量函數(shù)的聯(lián)合優(yōu)化效果。實(shí)驗(yàn)結(jié)果表明，所提出的算法在Auto Detected CUHK03 和VIPeR兩個(gè)數(shù)據(jù)庫上的準(zhǔn)確率均獲得顯著的提升，驗(yàn)證了所提出算法的優(yōu)越性。
-
關(guān)鍵詞:
- 行人再辨識(shí) /
- 深度學(xué)習(xí) /
- 三元卷積神經(jīng)網(wǎng)絡(luò)
Abstract: Most triplet Convolutional Neural Network (CNN) based person re-identification algorithms use the Euclidean distance as the similarity measurement between a pair of person images, and utilize the hinge loss function to train CNNs. However, there are two disadvantages in these approaches: the Euclidean distance is not discriminative enough for measuring person similarities; the margin parameter of the hinge loss function must be manually set in advance and it can not be adaptively adjusted. For these, a novel triplet convolutional neural network based person re-identification algorithm is proposed to solve the above two disadvantages for improving the accuracy. First, the normalization hybrid similarity function is proposed to replace Euclidean distance to obtain a more discriminative person similarity measurement. Second, the Log-logistic function is designed to replace the hinge function, which does not need to set the margin parameter so that the joint optimization effect of feature learning and similarity learning is improved. The experimental results on the Auto Detected CUHK03 and VIPeR databases show that the proposed method gains significant improvements in person re-identification accuracy, which verifies the superiority of the proposed method. -
GRAY Douglas and TAO Hai. Viewpoint invariant pedestrian recognition with an ensemble of localized features [C]. European Conference on Computer Vision, Marseille- France in Palais des Congrs Parc Chanot, 2008: 262-275. FARENZENA M, BAZZANI L, PERINA A, et al. Person re-identification by symmetry-driven accumulation of local features[C]. IEEE Conference on Computer Vision and Pattern Recognition, San Francisco, USA, 2010: 2360-2367. LIAO Shengcai and LI Stan Z. Efficient PSD constrained asymmetric metric learning for person re-identification[C]. IEEE International Conference on Computer Vision, Santiago, Chile, 2015: 3685-3693. MATSUKAWA Tetsu, OKABE Takahiro, SUZUKI Einoshin, et al. Hierarchical gaussian descriptor for person re- identification[C]. IEEE Conference on Computer Vision and Pattern Recognition, Las Vegas, USA, 2016: 1363-1372. CHEN Dapeng, YUAN Zejian, CHEN Badong, et al. Similarity learning with spatial constraints for person re-identification[C]. IEEE Conference on Computer Vision and Pattern Recognition, Las Vegas, USA, 2016: 1268-1277. YANG Xun, WANG Meng, HONG Richang, et al. Enhancing person re-identification in a self-trained subspace[OL]. https://arxiv.org/pdf/1704.06020, 2017. YANG Yang, WEN Longyin, LYU Siwei, et al. Unsupervised learning of multi-level descriptors for person re-identification [C]. AAAI Conference on Artificial Intelligence, San Francisco, California, USA, 2017: 4306-4312. WU Shangxuan, CHEN Ying Cong, LI Xiang, et al. An enhanced deep feature representation for person re-identification[C]. IEEE Winter Conference on Applications of Computer Vision, Lake Placid, NewYork, USA, 2016: 1-8. XIAO Tong, LI Hongsheng, OUYANG Wanli, et al. Learning deep feature representations with domain guided dropout for person re-identification[C]. IEEE Conference on Computer Vision and Pattern Recognition, Las Vegas, USA, 2016: 1249-1258. LI Wei, ZHAO Rui, XIAO Tong, et al. Deepreid: Deep filter pairing neural network for person re-identification [C]. IEEE Conference on Computer Vision and Pattern Recognition, Columbus, Ohio, USA, 2014: 152-159. YI Dong, LEI Zhen, LIAO Shengcai, et al. Deep metric learning for person re-identification[C]. International Conference on Pattern Recognition, Stockholm, Sweden, 2014: 34-39. VARIOR Rahul Rama, HALOI Mrinal, and WANG Gang. Gated siamese convolutional neural network architecture for human re-identification[C]. European Conference on Computer Vision, Amsterdam, Netherlands, 2016: 791-808. WU Lin, WANG Yang, LI Xue, et al. What-and-where to match: deep spatially multiplicative integration networks for person re-identification[OL]. https://arxiv.org/pdf/1707. 07074, 2017. ZHU Jianqing, ZENG Huanqiang, LIAO Shengcai, et al. Deep hybrid similarity learning for person re-identification[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2017, (99): 1. doi: 10.1109/TCSVT.2017. 2734740. CHEN S Z, GUO C C, and LAI J. Deep ranking for person re-identification via joint representation learning[J]. IEEE Transactions on Image Processing, 2016, 25(5): 2353-2367. doi: 10.1109/TIP.2016.2545929. ZHAO Liming, LI Xi, WANG Jingdong, et al. Deeply- learned part-aligned representations for person re- identification[OL]. https://arxiv.org/pdf/1707.07256, 2017. LIU H, FENG J, QI M, et al. End-to-end comparative attention networks for person re-identification[J]. IEEE Transactions on Image Processing, 2017, 26(7): 3492-3506. IOFFE Sergey and SZEGEDY Christian. Batch normalization: Accelerating deep network training by reducing internal covariate shift[C]. International Conference on Machine Learning, Lille, France, 2015: 448-456. KRIZHEVSKY Alex, SUTSKEVER Ilya, and HINTON Geoffrey E. ImageNet classification with deep convolutional neural networks[C]. International Conference on Neural Information Processing Systems, Lake Tahoe, Nevada, USA, 2012: 1097-1105. -
計(jì)量
- 文章訪問數(shù): 1703
- HTML全文瀏覽量: 176
- PDF下載量: 241
- 被引次數(shù): 0
下載:
