隨著老年人口比例增加，長期照護問題日益嚴重。一般而言，長照服務之照護者須提供長時間甚至全天候照料，容易在人事成本上造成負擔。市面上已有許多針對單一活動空間提供監測、追蹤年長者作息狀況的產品，但即使透過相機捕捉到該空間中當日的進、出門事件，仍然需要人為介入、標註每個事件中的人物身份後，才能獲得每位年長者當日的活動狀況，此過程繁瑣且需要花費大量標註人員之時間與精力。在本論文中，我們提出一套模組，能夠在獲得當日所有人物的移動軌跡 (Human Tracklet) 後，利用移動軌跡的特徵 (Feature) 以及分群 (Clustering) 的演算法，將所有移動軌跡進行分組；在人為介入時，只需檢查每一分組中的所有移動軌跡是否來自同一人，並將分群錯誤的組別中的移動軌跡重新進行人為標註，即可完成原先的任務。與原先需要逐一標註每段移動軌跡相比，使用本模組後可以節省不少標註的時間。此外，我們針對不同分群演算法與移動軌跡之特徵來源進行實驗，在分群結果的正確率上取得提升。最後，我們也驗證了此分群模組在內部資料集與兩個第三方資料集，在時間節省的比例上分別達到了平均 49.0%、35.3%、22.9%，此實驗結果表明，在不同環境下，我們提出的模組都能有效區分出不同人物的移動軌跡，並且確實減少了人為介入所需的時間。

As the proportion of elderly people increases, the problem of long-term care is growing. Generally speaking, the caregivers of long-term care services have to provide round-the-clock care, which can easily cause a burden of personnel costs. There are already many products which provide services of monitoring and tracking the physiological condition of elderly people in a single room, but even if the entrance and exit events are captured by a camera, it still needs human intervention to annotate the identity of the person in each event. This annotating process is tedious and requires a lot of time and effort of the annotator. In this paper, we propose a module that can group all the human tracklets after extracting their feature and performing clustering algorithms. Afterwards, it is only required to check the purity of each group and re-annotate the human tracklets in the impure groups for the human intervention. Comparing with the original process, this module can save a lot of time. In addition, we conduct experiments on different sources of feature of human tracklet and clustering algorithms and show that the accuracy of clustering improves. Finally, we further verify the effectiveness of the proposed module on an internal dataset and two public datasets, which achieves an improvement of 49.0%, 35.3% and 22.9%, respectively. The results show that the proposed module can effectively distinguish the identity of the person in each human tracklet under different scenes, and indeed reduce the time required for the human intervention.

1 緒論 1
2 相關文獻 5
2.1 Person Re-Identification 5
2.2 Long-Term Object Tracking 6
2.3 Clustering 7
3 研究方法 9
3.1 輸入與輸出 9
3.2 特徵來源 11
3.2.1 Re-ID 模型 11
3.2.2 Tracking 模型 13
3.2.3 再訓練 15
3.3 分群演算法 15
3.3.1 K-means Clustering 16
3.3.2 Agglomerative Clustering 16
3.4 評估指標 18
4 資料集 21
4.1 內部資料集 21
4.2 第三方資料集 23
4.2.1 DukeMTMC-VideoReID 23
4.2.2 MARS 23
5 實驗
5.1 實驗設定 27
5.2 內部資料集實驗結果 27
5.3 第三方資料集實驗結果 29
5.3.1 DukeMTMC-VideoReID 29
5.3.2 MARS 31
6 結論 33
References 35

[1] U. N. D. of Economic and S. Affairs, World Population Ageing 2019. United Nations, 2020.
[2] Y.-Y. Lin and C.-S. Huang, “Aging in Taiwan: Building a Society for Active Aging and Aging in Place,” The Gerontologist, vol. 56, pp. 176–183, 11 2015.
[3] “Fusion mmwave fall detection.” https://www.fusionnet.io/post/ fusioncare-mmwavefalldetection.
[4] “Privacy-preserving thermal image anomaly detection system.” https://www. chinatimes.com/realtimenews/20200730003897-260412.
[5] “Gigabyte fall detection.” https://www.gigabyte.com/tw/ Industry-Solutions/fall-detection.
[6] “Foreaider z smart sensing pad.” https://foreaider.com/en/1571-2/.
[7] “Cherry home healthcare solution.” https://get.cherryhome.ai/care/.
[8] E. Ristani, F. Solera, R. Zou, R. Cucchiara, and C. Tomasi, “Performance measures and a data set for multi-target, multi-camera tracking,” in European Conference on Computer Vision workshop on Benchmarking Multi-Target Tracking, 2016.
[9] Springer, MARS: A Video Benchmark for Large-Scale Person Re-identification, 2016.
[10] O. Oreifej, R. Mehran, and M. Shah, “Human identity recognition in aerial images,” in 2010 IEEE Computer Society Conference on Computer Vision and Pattern Recognition, pp. 709–716, 2010.
[11] K. Jüngling, C. Bodensteiner, and M. Arens, “Person re-identification in multi-camera networks,” in CVPR 2011 WORKSHOPS, pp. 55–61, 2011.
[12] R. Layne, T. Hospedales, and S. Gong, “Person re-identification by attributes,” vol. 2, 01 2012.
[13] W.-S. Zheng, S. Gong, and T. Xiang, “Reidentification by relative distance comparison,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 35, no. 3, pp. 653– 668, 2013.
[14] S. Pedagadi, J. Orwell, S. Velastin, and B. Boghossian, “Local fisher discriminant analysis for pedestrian re-identification,” in 2013 IEEE Conference on Computer Vision and Pattern Recognition, pp. 3318–3325, 2013.
[15] S. Liao, Y. Hu, X. Zhu, and S. Z. Li, “Person re-identification by local maximal occurrence representation and metric learning,” in 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 2197–2206, 2015.
[16] H. Liu, Z. Jie, J. Karlekar, M. Qi, J. Jiang, S. Yan, and J. Feng, “Video-based person re-identification with accumulative motion context,” CoRR, vol. abs/1701.00193, 2017.
[17] D. S. Bolme, J. R. Beveridge, B. A. Draper, and Y. M. Lui, “Visual object tracking using adaptive correlation filters,” in 2010 IEEE Computer Society Conference on Computer Vision and Pattern Recognition, pp. 2544–2550, 2010.
[18] J. Henriques, R. Caseiro, P. Martins, and J. Batista, “Exploiting the circulant structure of tracking-by-detection with kernels,” vol. 7575, pp. 702–715, 10 2012.
[19] J. F. Henriques, R. Caseiro, P. Martins, and J. Batista, “High-speed tracking with kernelized correlation filters,” CoRR, vol. abs/1404.7584, 2014.
[20] C. Ma, J.-B. Huang, X. Yang, and M.-H. Yang, “Hierarchical convolutional features for visual tracking,” in 2015 IEEE International Conference on Computer Vision (ICCV), pp. 3074–3082, 2015.
[21] G. Bhat, J. Johnander, M. Danelljan, F. S. Khan, and M. Felsberg, “Unveiling the power of deep tracking,” CoRR, vol. abs/1804.06833, 2018.
[22] M. Ester, H.-P. Kriegel, J. Sander, and X. Xu, “A density-based algorithm for discovering clusters in large spatial databases with noise,” in Proceedings of the Second International Conference on Knowledge Discovery and Data Mining, KDD’96, p. 226–231, AAAI Press, 1996.
[23] G. Wang, S. Gong, J. Cheng, and Z. Hou, “Faster person re-identification,” CoRR, vol. abs/2008.06826, 2020.
[24] J. Pang, L. Qiu, H. Chen, Q. Li, T. Darrell, and F. Yu, “Quasi-dense instance similarity learning,” CoRR, vol. abs/2006.06664, 2020.
[25] K. He, G. Gkioxari, P. Dollár, and R. B. Girshick, “Mask R-CNN,” CoRR, vol. abs/1703.06870, 2017.
[26] F. Yu, H. Chen, X. Wang, W. Xian, Y. Chen, F. Liu, V. Madhavan, and T. Darrell, “Bdd100k: A diverse driving dataset for heterogeneous multitask learning,” pp. 2633– 2642, 06 2020.
[27] Y. Wu, Y. Lin, X. Dong, Y. Yan, W. Ouyang, and Y. Yang, “Exploit the unknown gradually: One-shot video-based person re-identification by stepwise learning,” in The IEEE Conference on Computer Vision and Pattern Recognition (CVPR), June 2018.
[28] L. Zheng, L. Shen, L. Tian, S. Wang, J. Wang, and Q. Tian, “Scalable person reidentification: A benchmark,” pp. 1116–1124, 12 2015.