基於影像感測方法用於自動體外心臟電擊去顫器室內選址之人員時空分布估計

Image Sensing-Based Spatial-Temporal Demand Estimation for AED Installation in Buildings

邱文心(Wen-Xin Qiu)

指導教授 ： 陳柏華

國立臺灣大學/工學院/土木工程學研究所/碩士(2020年)

https://doi.org/10.6342/NTU202003934

摘要

各國推廣於公共場所室內安裝自動體外心臟電擊去顫器（AED）之政策已行之有年，其目的為提高心臟驟停患者之存活率。室內AED之選址已有許多既有之研究，然多採用數學規劃模型，而在現有模型中，利用人在室內的時空分布作為選址之依據，卻無較有效率之估計方法。因此本研究提出一估計室內需求分布之方法，作為室內AED選址之重要輸入參數。
為降低額外設備之建置成本，本研究提出之方法係利用一般常見之攝影機，並配合電腦視覺之影像偵測方法、相機模型之物像空間對應關係及資料探勘等方法，對於攝影機可見範圍之空間進行人數估計。本研究之輸入參數則包含室內路網、監視器影像及相機參數。本研究首先在監視影像畫面上偵測及追蹤各個人員之進出位置後，再加上對場景深度之估測並套用分群方法後，使用相機模型將實際座標與相機座標投影至同一空間以計算各節點上之人數。
本研究採用學校建物內的兩個實際案例來驗證所提出之方法，此二場景分別攝於長廊及大廳，以測試不同場景之表現並探討每個步驟所造成的誤差。最後輸出的總成果，在兩個案例中皆達到20%以下誤差。
綜上所述，本研究提出之以一般常見攝影機估測人員之時空分布的方法，其成果並能提供AED選址輸入參數之用。

關鍵字

自動體外心臟電擊去顫器 ； 人員偵測 ； 電腦視覺 ； 深度學習 ； 影像深度估計

並列摘要

Timely access of Automated External Defibrillator (AED) is crucial for increasing the survival rate of people with cardiac arrests. The optimum installation location of AEDs could be determined by mathematical programming models, while the spatial-temporal human distribution, the reference of where demands could happen, have not been estimated efficiently. This research aims to estimate the human distribution through image sensing. Inputs are assumed to be the in-building passage network, image sequences recording the traveling of human, and camera parameters. Human detection and tracking models are utilized to find humans in the images. Image depth estimation, clustering and the camera model are integrated for the matching of human and the in-building space in the image coordinates and real world coordinates. Finally, the temporal human count for each in-building space is acquired to be served as the input of AED location models.
To validate the approach, two real cases in a school building at corridor and hallway respectively are introduced to test the approach in different scenes, and a synthesized case is used to exclude error from detection and tracking step. The two cases both achieved less than 20% error for the final output.

並列關鍵字

Automated External Defibrillator ； Human Sensing ； Computer Vision ； Deep Learning ； Image Depth Estimation

參考文獻

[1] J. S. Bækgaard, S. Viereck, T. P. Møller, A. K. Ersbøll, F. Lippert, and F. Folke, "The Effects of Public Access Defibrillation on Survival After Out-of-Hospital Cardiac Arrest," Circulation, vol. 136, no. 10, pp. 954-965, 2017, doi: 10.1161/CIRCULATIONAHA.117.029067.

[2] American Heart Association, "Part 4: The Automated External Defibrillator," Circulation, vol. 102, no. suppl_1, 2000. [Online]. Available: https://www.ahajournals.org/doi/abs/10.1161/circ.102.suppl_1.I-60.

[3] 公共場所必要緊急救護設備管理辦法 [Public Emergency Medical Service Equipment Control Regulations], 2013.

[4] State Laws on Cardiac Arrest and Defibrillators - NCSL, USA, 2017.

[5] The Japan Emergency Medical Foundation, AEDの適正配置に関するガイドライン 補訂版 [The Guidelines for appropriate installations of automated external defibrillators (Revised Edition)], 2019. [Online]. Available: https://www.jhf.or.jp/check/upload_images/h30_aed_guidlines.pdf.