由圖9(a)的比較結果得,HOG檢測算法在旋轉角為(-10°,10°)區間內擁有高檢測率,但隨著旋轉角的增大,檢測率則不斷下降,而RGTHOG檢測算法則在測試旋轉角範圍內保持這良好的識別率。圖9(b)中HOG檢測算法誤檢率在[-45°,45°]範圍低於本文提出的檢測算法。
4 結語
本文提出的RGTHOG檢測算法采用經過RGT變換的具有旋轉不變性的梯度方向直方圖,將直方圖分割處理後將結果按扇區形式組合,並獲得RGTHOG特征描述子組,使用SVM算法建立兩級級聯分類器。通過第二級分類器的實驗數據表明單個RGTHOG特征描述子能有效刻畫出人體特征,在單一人體旋轉角的情況下的正檢測率達94%,人體旋轉角變化在 (-11.25°, 11.25°) 內的情況下的正檢測率不低於90%;RGTHOG特征描述子組在多個角度上具有旋轉不變性,檢測算法的實驗數據表明,待檢圖像在具有旋轉不變性的旋轉角上檢測效果最好。以上測試分類結果驗證了
RGTHOG檢測算法的可行性。使用線性核SVM作為級聯分類器的處理時間較長,對實時應用帶來一定影響,如何提高級聯分類器的處理速度仍有待研究。
參考文獻:
[1]LI X, GONG X, LI T, et al. Automatic pedestrian detection based on video surveillance [J]. Computer Science, 2014, 41(12): 264-268. (李新江,龔勳,李天瑞,等.麵向視頻監控的自動行人檢測[J].計算機科學,2014,41(12):264-268.)
[2]DOLLAR P, WOJEK C, SCHIELE B, et al. Pedestrian detection: an evaluation of the state of the art [J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2012, 34(4): 743-761.
[3]XU T, HUANG T, TIAN Y. Survey on pedestrian detection technology for onboard vision systems [J]. Journal of Image and Graphics, 2013, 18(4): 359-367. (許騰,黃鐵軍,田永鴻.車載視覺係統中的行人檢測技術綜述[J].中國圖象圖形學報,2013,18(4):359-367.)
[4]REILLY V, SOLMAZ B, SHAH M. Shadow Casting Out Of Plane (SCOOP) candidates for human and vehicle detection in aerial imagery [J]. International Journal of Computer Vision, 2013, 101(2): 350-366.
[5]ANDRILUKA M, SCHNITZSPAN P, MEYER J, et al. Vision based victim detection from unmanned aerial vehicles [C]// Proceedings of the 2000 IEEE/RSJ International Conference on Intelligent Robots and Systems. Piscataway: IEEE, 2010: 1740-1747.
[6]DALAL N, TRIGGS B. Histograms of oriented gradients for human detection [C] // CVPR 2005: Proceedings of the 2005 IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2005, 1: 886-893.
[7]MU Y, YAN S, LIU Y, et al. Discriminative local binary patterns for human detection in personal album [C] // CVPR 2008: Proceedings of the 2008 IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2008: 1-8.