基於室內人流影像辨識之自動體外心臟電擊去顫器設置決策

Video-based Indoor Human Detection for Decision-Making for the Installation Locations of Automated External Defibrillators

陳靖淳(Ching-Chun Chen)

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(2017年)

https://doi.org/10.6342/NTU201703955

摘要

本研究提出一偵測室內人流分佈的影像方法，目的為協助主管單位進行自動體外心臟去顫器設置選址之決策。自動體外心臟去顫器(Automated External Defibrillator，AED)為一重要之緊急救護裝置，主要針對心臟驟停之病患施行急救措施，提供經訓練之專業救護人員或未經訓練的一般民眾使用。其主要設置於人潮聚集之室內建築或室外地點。目前我國於現行法令中只列出符合特定條件之建物或是地點必須設置，卻未提及明確設置之規範細節。因此本研究期望透過偵測人流影像的方式，蒐集建物室內人群的實際分佈，用以協助進行設置位置之最佳化決策。研究方法中我們錄製建物內部行人的影片，之後藉由影像處理的程序進行人員的影像偵測。在行人偵測的部分，本研究以方向梯度直方圖 (Histogram of Oriented Gradient，HOG) 和Oriented FAST and Rotated BRIEF (ORB) 作為特徵描述子，並結合支持向量機 (Support Vector Machine，SVM) 作為分類器，藉以獲得一段時間內人員於該建物內部的分佈情形，進而將該分佈資料結合目標建物之室內路網模型，分別作為參數資料導入最佳化位置模型，進行選址問題之求解，得出目標建物的自動體外心臟去顫器最佳設置位置建議，並提升自動體外心臟去顫器於緊急狀況發生時的效益。

關鍵字

自動體外心臟電擊去顫器 ； 行人偵測 ； 影像處理 ； 方向梯度直方圖 ； 支持向量機

並列摘要

A vision based method was introduced to detect human patterns for in-door environments. The purpose of the research is to assist decision makers at which locations should automated external defibrillators (AEDs) be installed. AED is a critical first-aid equipment which can be used by professionals or bystanders to give the first aid treatment to the sudden cardiac victims. It is mainly installed at the crowded outdoor public places or buildings. Currently, the regulation in Taiwan only regulate buildings and places that meet specific conditions should install AEDs. It does not offer specific details for people to follow while installing AEDs. Therefore, the study observes the actual distribution of pedestrian in the building by human detection techniques. These data would be used as an important parameter while calculating the optimized install locations. We first recorded the video of people in the target building. During the process of detection, we adopt the Histogram of Oriented Gradient (HOG) and the Oriented FAST and Rotated BRIEF (ORB) as the feature descriptor. Additionally, using Support Vector Machine (SVM) as the classifier to classify the target and detect the locations of people at each time interval. After the extraction of the human volume data, we consider it as a critical parameter in the decision making model. The distribution data we collected previously and the distance data we extracted from the geometric network would be used in the decision making analysis. The analysis would help to plan and evaluate locations of AEDs based on the real potential demand distribution and could optimize the performance of AED during an emergency.

並列關鍵字

Automated external defibrillators ； human detection ； image processing ； histogram of oriented gradients ； support vector machine

參考文獻

Balasubramanian, P. (2016). “Improving Gradient Histogram Based Descriptors for Pedestrian Detection in Datasets with Large Variations.” Cvpr, 104–113.

Bradley, A. P. (1997). “The use of the area under the ROC curve in the evaluation of machine learning algorithms.” Pattern Recognition, 30(7), 1145–1159.

Cai, Z., Saberian, M., and Vasconcelos, N. (2015). “Learning Complexity-Aware Cascades for Deep Pedestrian Detection.” Iccv, 3361–3369.

Chan, T. C. Y., Demirtas, D., and Kwon, R. H. (2016). “Optimizing the Deployment of Public Access Defibrillators.” Management Science, 62(12), 3617–3635.

Chen, A. Y., and Huang, T. (2015). “Toward BIM-Enabled Decision Making for In-Building Response Missions.” IEEE Transactions on Intelligent Transportation Systems, 16(5), 2765–2773.