计算机工程与应用
計算機工程與應用
계산궤공정여응용
COMPUTER ENGINEERING AND APPLICATIONS
2013年
10期
159-162
,共4页
孙水发%覃音诗%马先兵%雷帮军
孫水髮%覃音詩%馬先兵%雷幫軍
손수발%담음시%마선병%뢰방군
前景检测%ViBe算法%形态学处理%ROC曲线
前景檢測%ViBe算法%形態學處理%ROC麯線
전경검측%ViBe산법%형태학처리%ROC곡선
foreground detection%ViBe algorithm%morphology post-processing%ROC curve
背景差分法是实际中应用最广泛的前景检测方法,其关键是背景建模,比较常用的背景建模方法是高斯混合模型GMM(Gaussian Mixture Model).最近一种称为视觉背景抽取算法ViBe(Visual Background extractor)由于其简单、快速的特点得到了越来越多的重视.但对于存在动态背景的户外视频,仍然存在噪声及背景的干扰.提出用形态学方法对算法进行改进,即先用开操作来消除噪声,再用闭操作来填充物体内细小空洞等.用ROC曲线测试了算法性能,结果表明,进行形态学处理后算法性能有了比较大的提高,比如对于户外视频Watersurface,在FPR为1%时TPR最高提高了31%.
揹景差分法是實際中應用最廣汎的前景檢測方法,其關鍵是揹景建模,比較常用的揹景建模方法是高斯混閤模型GMM(Gaussian Mixture Model).最近一種稱為視覺揹景抽取算法ViBe(Visual Background extractor)由于其簡單、快速的特點得到瞭越來越多的重視.但對于存在動態揹景的戶外視頻,仍然存在譟聲及揹景的榦擾.提齣用形態學方法對算法進行改進,即先用開操作來消除譟聲,再用閉操作來填充物體內細小空洞等.用ROC麯線測試瞭算法性能,結果錶明,進行形態學處理後算法性能有瞭比較大的提高,比如對于戶外視頻Watersurface,在FPR為1%時TPR最高提高瞭31%.
배경차분법시실제중응용최엄범적전경검측방법,기관건시배경건모,비교상용적배경건모방법시고사혼합모형GMM(Gaussian Mixture Model).최근일충칭위시각배경추취산법ViBe(Visual Background extractor)유우기간단、쾌속적특점득도료월래월다적중시.단대우존재동태배경적호외시빈,잉연존재조성급배경적간우.제출용형태학방법대산법진행개진,즉선용개조작래소제조성,재용폐조작래전충물체내세소공동등.용ROC곡선측시료산법성능,결과표명,진행형태학처리후산법성능유료비교대적제고,비여대우호외시빈Watersurface,재FPR위1%시TPR최고제고료31%.
Background subtraction is a foreground detection method, and the key step of which is the background modeling. GMM(Gaussian Mixture Model)background modeling method is used most widely. Recently a method called ViBe(Visual Background extractor)has caught more and more attention for its simplicity and high speed. It’s still difficult to detect the fore-ground for the outdoor video because of the the complex background, such as the shivering or trembling leaves of the trees. Regarding this fact, a morphology post-processing is used to filter the results of ViBe, i.e. with the opening operation to remove the noise, then with the closing operation to fill small voids within the object. The performance is evaluated with ROC curve. The results show that the performance is improved greatly with the morphology post-processing, for outdoor video Watersurface, TPR of the system performance is improved 31%after morphology processing when FPR is 1%.