臺灣博碩士論文加值系統

Single-shot person re-identification is to match pedestrian images from different cameras under the condition of large illumination variations, different viewpoints, cluttered background, and inadequate information of single-shot case. To deal with these challenges, we propose a single-shot person re-identification method with improved pedestrian segmentation by incorporating shape prior and color seed constraints into Random Walk in this thesis. Our method is a four-step procedure which includes finding feature correspondences and person recognition. Based on an improved Random Walks algorithm, we segment accurate foreground masks from a detected pedestrian region by combining the shape prior information and the color histogram constraint into the Random Walk formulation. . Then color features of HSV histogram and 1-D RGB signal aided by texture feature from each part are used for the person recognition. In addition, we choose the correct match by matching the similarity scores of all features with appropriate weight selection. The experimental results demonstrate improved segmentation results on some real images by using the modified Random walk segmentation. In addition, we also show superior performance on person re-identification accuracy by using the proposed algorithm compared to the previous representative methods.

單張影像行人辨識是想在亮度差異大、不同視角且背景複雜的狀況加上單張影像不足夠的資訊困難中去找尋出現於不同位置上的行人。在本文中我們加入人行機率分佈及初始點上的顏色資訊來改善隨機漫步行人影像分割，並提出基於改善後的行人影像分割實做單張影像行人辨識去處理前面提及的困難。我們所提出的方法是個四步驟的流程，其中又可分成兩大部分:定義相對區域及行人辨識。根據改善後的隨機漫步影像分割，我們可以在行人偵測區域中取得更精確的前景。接著是從相對應的區塊中獲得色相飽和度明度分佈及一維三原色光訊號的顏色特徵，同時加上紋理分析的幫助做行人辨識。除此之外，我們利用自己定義的權重篩選來合併三種特徵決定出影像比對的相似值再去找尋正確的行人配對。透過實驗結果顯示改善後的隨機漫步影像分割提升了真實影像上前景區塊的結果。同時利用我們的行人辨識方法所得到的準確率也比過去代表性的方法有卓越的表現。

Contents i
List of Figures ii
List of Tables iii
Chapter 1. Introduction 1
1.1. Problem Description and Motivation 1
1.2. System Overview 2
1.3. Main Contribution 4
1.4. Thesis Outline 5
Chapter 2. Related Works 6
2.1. Extracting distinctive features with correspondences 6
2.2. Learning discriminatory feature representation 9
Chapter 3. Proposed Method 13
3.1. Pedestrian Segmentation 13
3.1.1. Human shape prior estimation 14
3.1.2. Generative Model for Segmentation 15
3.2. Human Region Partitioning 23
3.3. Signature Extraction and Representation 26
3.3.1. Color histogram on HSV color space 26
3.3.2. 1-D signal on RGB color space 28
3.3.3. LBP and SILTP as texture feature 29
3.4. Feature matching for ranking score 31
Chapter 4. Experimental Results 34
4.1. Pedestrian segmentation performance 34
4.2. Person re-identification performance 37
Chapter 5. Conclusion 44
Bibliography 46

[1] N. Gheissari, T. B. Sebastian, P. H. Tu, J. Rittscher, and R. Hartley. Person reidentification using spatiotemporal appearance. In CVPR, vol. 2, pages 1528–1535, 2006.
[2] M. Farenzena, L. Bazzani, A. Perina, M. Cristani, and V. Murino. Person re-identification by symmetry-driven accumulation of local features. In CVPR, 2010.
[3] O. Hamdoun, F. Moutarde, B. Stanciulescu, and B. Steux. Person re-identification in multi-camera system by signature based on interest point descriptors collected on short video sequences. In Proceedings of the IEEE Conference on Dis- tributed Smart Cameras, pages 1–6, 2008.
[4] N. Bird, O. Masoud, N. Papanikolopoulos, and A. Isaacs. Detection of loitering individuals in public transportation areas. Intelligent Transportation Systems, IEEE Transactions on, 6(2):167–177, June 2005.
[5] L. Bazzani, M. Cristani, A.Perina, M. Farenzena, and V. Murino. Multiple-shot person re-identification by HPE signature. In Proceedings of 20th International Conference on Pattern Recognition, ICPR 2010
[6] W. Schwartz and L. Davis. Learning discriminative appearance-based models using partial least squares. In XXII SIBGRAPI, 2009.
[7] D. Gray and H. Tao. Viewpoint invariant pedestrian recognition with an ensamble of localized features. In ECCV, pages 262–275,2008.
[8] Z. Lin and L. Davis. Learning pairwise dissimilarity profiles for appearance recognition in visual surveillance. In ISVC’08: Proceedings of the 4th International Symposium on Advances in Visual Computing, pages 23–34, 2008.
[9] K-C. Li and S-H. Lai. Automatic pedestrian image segmentation by using human shape prior.
[10] Inmar E. Givoni and Brendan J. Frey. A Binary Variable Model for Affinity Propagation. Neural Computation, Vol. 21, issue 6, pp 1589-1600, June 2009.
[11] L. Grady. Multilabel RandomWalker Image Segmentation Using Prior Models. In CVPR, Vol. 1, pp. 763-770, 2005.
[12] L. Grady. Random walks for image segmentation. In PAMI, 28(11):1768–1783, 2006.
[13] ViSOR Video Surveillance Online Repository.
http://www.openvisor.org/.
[14] MIT-CBCL Pedestrian Dataset. http://cbcl.mit.edu/cbcl/software-datasets/PedestrianData .html.
[15] D. Gray, S. Brennan, and H. Tao. Evaluating appearance models for recognition, reacquisition and tracking. In PETS, 2007.
The available site for VIPeR dataset .
http://vision.soe.ucsc.edu/?q=node/178 .
[16] T. Ojala, M. Pietik¨ainen, and M. M¨aenp¨a¨a. Multiresolution gray-scale and rotation invariant texture classification with local binary patterns. IEEE Transactions on Pattern Analysis and Machine Intelligence, 24(7):971–987, 2002.
[17] S. Liao, G. Zhao, V. Kellokumpu, M. Pietikäinen, and S. Z. Li. Modeling Pixel Process with Scale Invariant Local Patterns for Background Subtraction in Complex Scenes,” In Proceedings of IEEE Computer Society Conference on Computer Vision and Pattern Recognition. June 13-18, 2010.
[18] W-S. Zheng, S. Gong, and T. Xiang. Person Re-identification by Probabilistic Relative Distance Comparison. In CVPR, 2011.
[19] N. Jojic, A. Perina, M. Cristani, V. Murino, and B. Frey. Stel component analysis: Modeling spatial correlations in image class structure. In CVPR, pages 2044–2051.