本論文對於人流計數提出了一個新的方法，利用橢圓偵測和HOG上半身偵測搭配HMM和追蹤來實現。首先前景的物件剪影會被萃取出來稱作blobs，藉由分析blob的資訊然後基於行人和行人之間的關係而產生blobs之間的鏈結。由鏈結所得的資訊可得知此物件blob在偵測畫面區域中的過往訊息，如此可得知物件是否和別的物件發生碰撞或是物件維持著狀態的時間等。為了計算出blob中的物件，我們使用橢圓偵測以及HOG上半身偵測以尋找物件個數。使用橢圓偵測是以橢圓逼近行人的輪廓並快速尋找匹配的blob面積，而使用HOG偵測是確保當blob資訊不夠完整時，能夠直接在畫面中搜尋行人的位置。在而為了解決重疊的問題，我們使用HMM用以檢測各種不同的鏈結上所記錄的blob狀態而找出最符合的假設再加以判別人數，和以往方法最大的不同點是，我們是使用物件從進入場景到離開場景這整段時間的所有blob資訊作為人流計數的依據，相較於只考慮物件幾張特定畫面做分析的方法會來的更精確嚴謹。 在最後的實驗中，由實驗數據可以證明本論文所提出方法的可行性。

This paper presents a new people counting approach using ellipse detection , HOG upper body detection, HMM and tracking. First of all, the foreground object silhouettes are extracted described as blobs. The linkage is generated by analyzing the blob information between blobs and the relationship between pedestrian. We can get the object blobs previous states by analyzing the linkage information in the region of interest. Then we have the information of the objects which are merged or separated. To count the objects in the blob, we use ellipse detection and HOG upper body detection to get the number of objects in the blob. We use ellipse detection by matching the area and fitting the outline of the blobs with the ellipse, and then we use HOG upper body detection to detect the position of the pedestrian if the blob information is not enough. To solve occlusion problem, we use HMM to model the variations blob states on each linkage, which can be used to find the most matching hypothesis to determine the people number. Different from previous methods, we analyze the image sequences from the objects entering the scene till the objects exiting the scene. Compare with the specific frames analyzing way, we have better accuracy. In the experiments, we illustrate the effectiveness of our method.

[1] Y. Benabbas, et al.“ Spatio-Temporal Optical Analysis for People Counting,” IEEE AVSS, 2010.
[2] D. Ryan, et al., “Scene Invariant Crowd Counting for Real-time Surveillance,” ICSPCS 2008.
[3] D. Ryan, et al., “Crowd Counting using Multiple Local Features,” IEEE Digital Image Computing: Techniques and Applications, 2009.
[4] D. Ryan, et al., “Crowd Counting using Group Tracking and Local Features,” IEEE AVSS, 2010.
[5] A. Chan, et al., “Privacy Preserving Crowd Monitoring: Counting People without People Models or Tracking,” IEEE CVPR 2008.
[6] D. Fehr, et al.,“Counting People in Groups,” IEEE AVSS, 2009.
[7] Y. L. Hou, et al., “People Counting and Human Detection in a Challenging Situation,” IEEE Trans. on SMC Part A, Vol. 41, No. 1, Jan. 2011.
[8] O Sidla, et al., “Pedestrian Detection and Tracking for Counting Applications in Crowded Situation,” IEEE AVSS, 2006
[9] C. Zeng et al., “Robust Head-Shoulder Detection by PCA-Based Multilevel HOG-LBP Detector for People Counting,” IEEE ICPR 2010.
[10] M. Lin, et al., “Estimating the Number of People in Crowded Scenes by MID Based Foreground Segmentation and Head-Shoulder Detection,” IEEE CVPR 2008.
[11] V. Rabaud et al.“Counting Crowded Moving Objects,” IEEE CVPR 2006.
[12] Y. Cong, et al., “Flow Mosaicking: Real-Time Pedestrian Counting without Scene-specific Learning,” IEEE CVPR 2009.
[13] G. Antonini, et al., “Counting Pedestrians in Video Sequences using Trajectory Clustering,” IEEE Trans. on CAS for VT, 16(8), Aug. 2006.
[14] K. Kim, et al., “Real-Time Foreground- Background Segmentation Using Codebook Model,” Real-Time Imag. 11(3), 172~185, 2005.
[15] A. Ilyas, et al., “Real-Time Foreground- Background Segmentation Using a Modified Codebook Model,” IEEE AVSS, 2009.
[16] T. Zhao and R. Nevatia, “Bayesian human segmentation in crowded situations,” Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition, 2003., vol. 2, pp. II–459–66 vol.2, June 2003.
[17] T. Zhao and R. Nevatia, “Tracking multiple humans in crowded environment,” Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition, 2004., vol. 2, pp. II–406–II–413 Vol.2, June-2 July 2004.
[18] D. Kong, D. Gray, and H. Tao, “Counting pedestrians in crowds using viewpoint invariant training,” in 18th International Conference on Pattern Recognition, ICPR, pp. 1187–1190, 2006.
[19] 孫伊廷 “混合式編碼簿模型於監視系統應用” 國立清華大學 碩士論文2010
[20] Chung-Lin Huang, Shih-Chung Hsu, I-Chung Tsao, Ben-Syuan Huang “People Counting using Ellipse Detection and Tracking,” Asian Conference on Pattern Recognition, ACPR, 2011.
[21] N. Dalal, B. Triggs, “Histograms of oriented gradients for human detection,” IEEE Conf. Computer Vision and Pattern Recognition, vol. 1, pp. 886-893, July 2005.