人物過近機制濾除人物假象

觀察從第 12534 幀﹙無假象人物，如圖 5-3(a)﹚到第 12537 幀的結果，其中人 物之初始數量雖然在後三個幀中有出現第十個人物﹙假象﹚的情形﹙如圖 5-3(b)-(d)﹚，

但第十人透過人物過近機制被濾除掉，並且經過體積的判斷與權重調整位置，可 看出在場景中九人之相關位置依然能被準確定位﹙如圖 5-4(a)-(d)黃圈之結果﹚。

詳細數據請見表格 1。

(a) (b)

(c) (d)

圖 5-3 (a)黃圈為無假象之定位結果。(b)-(d)中黃圈顯示修正過後人物均能被穩定定位

(a) (b)

(c) (d)

圖 5-4 (a)-(d)分別為圖 5-3 (a)-(d)之實際定位結果

圖 5-5 場景 4 中第 609 幀的結果

表格 1 場景 1 至場景 5 中的正確率與執行速度之量化結果。

Frames

數量 正確偵測 Miss detection False alarm Recall Precision Error Avg.

﹙cm﹚

Error Std.

﹙cm﹚ FPS Avg.

S1 690 5967 252 274 0.959479 0.956097 11.55673 39.55099 19.24 S2 775 5764 320 408 0.947403 0.933895 11.19735 41.03482 18.31 S3 270 3032 220 345 0.932349 0.897838 11.78517 42.07284 12.31 S4 70 420 51 45 0.89172 0.903226 10.38212 32.62293 10.31 S5 40 234 46 20 0.835714 0.92126 13.08267 61.93023 10.83

六、 結論

本篇論文利用人物佔地的概念來進行研究，其中僅使用了前景資訊，在不分 析人物特徵點的情況下，除了利用軸線概念外，還使用了﹙1﹚人物候選區域之地 面均勻取樣﹙一維 USGP﹚方式，來減少在投影軸線上的計算成本；﹙2﹚在一維 USGP 上隨機取樣產生人物的初始位置與人數：在人物候選的位置取樣上，使用隨 機取樣與檢查距離的方式大量地減少計算 IFR 的開銷；﹙3﹚體積概念依權重來定 位，以獲得穩定的人物定位：經過隨機取樣得到初始位置與人數後，重新的區域 取樣﹙二維 USGP﹚使得體積與人物過近機制能將假象人物妥善地被濾除；﹙4﹚

體積權重來重新衡量人物位置，能保持人物定位的準確度。本研究有效率地大量 降低系統的計算成本，符合即時運算的需求，並且在人物靠近的情形下﹙大於 60 cm﹚，依然能不失定位的精準度。

參考文獻

〔1〕Luo Sun, Huijun Di, Linmi Tao, Guangyou Xu, "A Robust Approach for Person Localization in Multi-Camera Environment, " in Proceedings of International Conference on Pattern Recognition, pp. 4036-4039, 2010.

〔2〕Murtaza Taj, Andrea Cavallaro, "Multi-View Multi-Object Detection and Tracking," in Collection of Computer Vision: Detection, Recognition and Reconstruction, pp. 263-280, 2010.

〔3〕Kuo-Hua Lo and Jen-Hui Chuang, "Vanishing Point-Based Line Sampling for Efficient Axis-Based People Localization," IEEE International Conference on Image Processing, pp. 11-14, Sep. 2011.

〔4〕Jamal Saboune and Robert Laganiere, "People Detection and Tracking Using the Explorative Particle Filtering," IEEE International Conference on Computer Vision, pp.1298-1305, 2009.

〔5〕K. Kim and L. Davis, "Multi-Camera Tracking and Segmentation of Occluded People on Ground Plane Using Search-Guided Particle Filtering," in Proceeding of European Conference on Computer Vision, pp. 98-109, 2006.

〔6〕Weina Ge and Robert T. Collins, "Crowd Density Analysis with Marked Point Processes," IEEE Signal Processing Magazine, pp. 107-111 and pp. 123, Sep.

2010.

〔7〕Weina Ge and Robert T. Collins, "Crowd Detection with a Multiview Sampler,"

European Conference on Computer Vision, pp. 324-337, 2010.

〔8〕László Havasi and Zoltán Szlávik, "Using Location and Motion Statistics for the Localization of Moving Objects in Multiple Camera Surveillance Videos," IEEE International Workshop on Visual Surveillance, pp. 1275-1281, 2009.

Conference on Computer Vision, pp.133-146, 2006.

〔10〕Anastasios L. Kesidis, Dimitrios I. Kosmopoulos, "Robust Occlusion Handling with Multiple Cameras Using a Homography Constraint," in Proceedings of International Conference on Computer Vision Theory and Applications, pp.

560-565, 2009.

〔11〕Santos, T. T. and Morimoto, C. H., "Multiple Camera People Detection and Tracking Using Support Integration," Pattern Recognition Letters, vol. 32, no. 1, pp. 47-55, 2011.

〔12〕R. Eshel and Y. Moses, "Homography Based Multiple Camera Detection and Tracking of People in a Dense Crowd," IEEE Conference on Computer Vision and Pattern Recognition, pp.1-8, 2008.

〔13〕Weiming Hu, Min Hu, Xue Zhou, Tieniu Tan, Jianguang Lou and Steve Maybank, "Principal Axis-Based Correspondence between Multiple Cameras for People Tracking," IEEE Transactions on Pattern Analysis and Machine

Intelligence, vol. 28, no. 4, pp. 663-671, Apr. 2006.

〔14〕Horng-Horng Lin, Jen-Hui Chuang and Tyng-Luh Liu, " Regularized Background Adaptation: A Novel Learning Rate Control Scheme for Gaussian Mixture Modeling," IEEE Transactions on Image Processing, vol. 20, no. 3, pp.

822-836, Mar. 2011.