李義超+羅飛
收稿日期:2013-11-11
作者簡介:李義超(1988—),男,湖北漢川人,碩士研究生,研究方向:數字圖像處理,運動目標的識別與跟蹤。
通訊聯系人,E-mail:liyichao-168@163.com
文章編號:1003-6199(2014)03-0088-04
摘 要:提出一種改進的適用于智能安防領域中離崗檢測的目標跟蹤算法,該算法結合均值漂移算法和粒子濾波算法的優點,先使用均值漂移算法對目標進行預跟蹤,然后在此基礎上使用粒子濾波對目標精確定位,在保證了跟蹤準確率的前提下縮短了算法的計算時間。此外,針對監控視頻大多分辨率低,目標辨識度不高等特點,在本文中,原始視頻流的灰度信息和紋理信息被作為待跟蹤目標的特征。實驗結果證明,采用該混合特征的目標跟蹤算法比其他同類算法在目標跟蹤的準確率和實時性上具有更好的表現,能夠適應更廣泛的視頻場景。
關鍵詞:目標跟蹤;均值漂移;粒子濾波;混合特征;離崗檢測
中圖分類號:TP391.9 文獻標識碼:A
An Object Tracing Algorithm Suitable for Off-position Detection
LI Yi-chao, LUO Fei
(Automation Science and Engineer Academy, South China University of Technology,Guangzhou,Guangdong 510641,China)
Abstract:In this paper, we propose an improved object tracing algorithm which is suitable for off-position detection in intelligent security filed. This algorithm takes advantage of mean shift and particle filter, pre-traces the object by the mean-shift algorithm and then calculates the accurate position by particle algorithm, which shortens computing time on the premise of insuring tracing accuracy. Besides, according to the problem that the low-resolution and low contrast of surveillance video, a new hybrid feature based on the gray scale information and texture information is regarded as the main feature of object in video scene. At last, experiment results prove that the improved object tracing algorithm with hybrid feature have better performance of tracking accuracy and real-time, which is applied more widely than other algorithms.
Key words:object tracing; mean-shift; particle filter; hybrid feature; off-position detection
1 引 言
離崗檢測是運用目標跟蹤算法對視頻場景中的執勤人員進行跟蹤,它是智能安防領域中智能視頻分析的一個重要的研究方向,具有廣泛的實際應用價值。目前,國內外的研究人員針對視頻場景中特定的目標進行跟蹤的課題已經提出了許多目標跟蹤算法,如均值漂移(mean-shift)算法、粒子濾波算法等,但是卻沒有一種有效的算法專門應用于離崗檢測。
文獻[1],[2],[3],[4]提出了幾種粒子濾波算法,不過這些算法均難以應用于安防領域內的視頻場景。文獻[8]針對目標是否被遮擋分別采用均值漂移算法與粒子濾波算法,但是從本質上來講,該算法并沒有將二者融合,只是根據不同的視頻場景選擇不同的算法。……