利用HOG於影像基礎之行人偵測於室內緊急應變

Using HOG for Video-based Human Detection for In-building Emergency Response

張正誼(Cheng-Yi Chang)

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(2015年)

https://doi.org/10.6342/NTU.2015.01362

摘要

行人偵測及追蹤在智慧型運輸系統當中是一項重要的應用。行人流量的資料能幫助推估通道容量，而蒐集行人軌跡能夠從中觀測一些行人在空間中的移動傾向。本研究呈現基於影像之自動化行人追蹤及計算方法，預期將應用於室內通道之特性擷取及緊急救災時之情蒐。研究方法利用方向梯度直方圖(Histogram of Oriented Gradient)，作為訓練樣本的特徵，進而訓練支持向量機(Support Vector Machine)作為分類器。結合影像去背法(Background Subtraction)，本研究採用之計算流程能夠快速且準確偵測行人。而卡爾曼濾波器(Kalman Filter)及匈牙利演算法(Hungarian Algorithm)則於計算流程中幫助追蹤並建立行人軌跡。本研究最後針對數個行人樣本進行驗證，以瞭解本研究所提出之系統效能。結果顯示本研究的自動化行人追蹤及計算系統有潛力應用於室內緊急應變。

關鍵字

行人偵測 ； 方向梯度直方圖 ； 多目標行人追蹤 ； 行人計算 ； 卡爾曼濾波 ； 支持向量機 ； 影像去背

並列摘要

Human detection and tracking is an important application in intelligent transportation systems. The data of pedestrian flow could help estimate passage capacity, and collection of pedestrian trajectories could observe the movement tendencies of people passing through the area. In this thesis, we present an approach for video-based automatic people tracking and counting. This approach is expected to be utilized in in-building emergency situations. By extracting the Histograms of Oriented Gradient (HOG) features from the training dataset, the Support Vector Machine (SVM) is trained as the human detector. By combining the HOG detector with Background Subtraction, the detection process could achieve rapid and accurate detections. The Kalman Filter and the Hungarian Algorithm are utilized in this process for the establishment of human trajectory tracking. We test the system on several pedestrian datasets to validate the performance of the proposed system. The automated pedestrian counting system has shown its potential provided the tracking results.

並列關鍵字

Human Detections ； HOG ； Multi-human Tracking ； Human Countin ； Kalman Filter ； SVM classifier ； Background Subtraction

參考文獻

1. Luh, P.B., et al., Modeling and optimization of building emergency evacuation considering blocking effects on crowd movement. Automation Science and Engineering, IEEE Transactions on, 2012. 9(4): p. 687-700.

2. Lv, W., et al., A Two-Dimensional Optimal Velocity Model for Unidirectional Pedestrian Flow Based on Pedestrian's Visual Hindrance Field. Intelligent Transportation Systems, IEEE Transactions on, 2013. 14(4): p. 1753-1763.

3. Chen, A.Y. and T. Huang, Toward BIM-Enabled Decision Making for In-Building Response Missions. (Accepted) Intelligent Transportation Systems, IEEE Transactions on, 2015.

6. Lowe, D.G. Object recognition from local scale-invariant features. in Computer vision, 1999. The Proceedings of The Seventh IEEE International Conference on. 1999. IEEE.

7. Chen, P.-Y., et al., An efficient hardware implementation of HOG feature extraction for human detection. Intelligent Transportation Systems, IEEE Transactions on, 2014. 15(2): p. 656-662.