【摘要】：隨著標準化考場建設(shè)的推進,傳統(tǒng)視頻監(jiān)控技術(shù)監(jiān)控效率低、海量視頻存儲壓力大的缺點逐漸突顯出來。智能化監(jiān)考系統(tǒng)是智能行為分析技術(shù)的一個應(yīng)用方面,可以杜絕監(jiān)考不嚴,提高監(jiān)控效率,緩解視頻存儲壓力。因此,基于考場視頻監(jiān)控的智能監(jiān)考方法的研究具有非常重要的實際意義,不僅減小了人力財力物力的投入,也提高了考試公平性。本文在考場環(huán)境下提出了一套智能監(jiān)考方法,適用于考生出勤情況記錄和對考生異常行為進行智能檢測。本文工作將從以下四個方面展開:(1)通過對考生坐姿特點的觀察,提出基于考生頭肩部位的目標檢測方法。結(jié)合方向梯度直方圖特征和等價局部二值模式直方圖特征,構(gòu)建融合特征。采用支持向量機分別結(jié)合單一特征和融合特征訓(xùn)練分類器,在考生實驗數(shù)據(jù)集上進行目標檢測實驗并分析檢測性能。提出采用基于分類器級聯(lián)的考生檢測框架,以滿足檢測率和檢測速度的雙重要求。(2)考慮到考場環(huán)境的特殊性,提出了基于YCbCr顏色空間的膚色、發(fā)色檢測和隨機抽樣一致性的誤差處理方法來修正考生檢測結(jié)果,以達到考生人數(shù)統(tǒng)計和出勤情況記錄的目的。(3)提出了基于稀疏重建的考生異常行為檢測方法,采用時空梯度特征描述考生行為的外觀特征,通過提取運動關(guān)注區(qū)域和主成分分析的方式簡化原始樣本數(shù)據(jù)以減少計算量。對考生常規(guī)行為樣本數(shù)據(jù)進行稀疏組合學(xué)習(xí)并建立模型,通過該模型對每個測試樣本計算相應(yīng)的重建誤差,以此完成考生異常行為檢測。在本文實驗數(shù)據(jù)集上,該方法可以取得較高的檢測性能,并且可以達到實時檢測的速度。(4)針對基于稀疏重建的考生異常行為檢測方法性能上的不足,本文增加了基于運動歷史圖像的運動連通域檢測方法,形成基于多信息融合的雙通道檢測框架。雙通道下,考生異常行為的檢測性能得到有效提升。將本文方法與其它常見的考生可疑行為檢測方法進行對比和分析,通過實驗證明了本文提出的方法具有更好的普適性。
[Abstract]:With the development of standardized examination room, the shortcomings of traditional video surveillance technology, such as low efficiency and high pressure of mass video storage, are becoming more and more obvious. Intelligent invigilation system is an application aspect of intelligent behavior analysis technology. It can eliminate invigilation, improve the efficiency of monitoring and relieve the pressure of video storage. Therefore, the research of intelligent invigilation method based on video surveillance has very important practical significance, which not only reduces the investment of manpower, finance and material resources, but also improves the fairness of examination. In this paper, an intelligent invigilation method is proposed, which can be used to record the attendance of examinees and to detect the abnormal behavior of examinees. The work of this paper will be carried out from the following four aspects: (1) by observing the characteristics of sitting posture of examinees, a target detection method based on the head and shoulder of examinees is proposed. Combining the directional gradient histogram feature and the equivalent local binary pattern histogram feature, the fusion feature is constructed. Using support vector machine (SVM) to train classifier with single feature and fusion feature respectively, the target detection experiment was carried out on the data set of examinee experiment and the detection performance was analyzed. The framework of examinee detection based on classifier concatenation is proposed to meet the double requirements of detection rate and speed. (2) considering the particularity of examination environment, the color of skin based on YCbCr color space is proposed. The error processing method of color detection and random sampling consistency is used to correct the test results in order to achieve the purpose of examinee number statistics and attendance record. (3) A sparse reconstruction-based method for examinee abnormal behavior detection is proposed. The spatio-temporal gradient features are used to describe the appearance features of examinees' behavior. The original sample data is simplified by extracting the region of motion concern and principal component analysis to reduce the computational complexity. The sample data of routine behavior of examinee is studied by sparse combinatorial learning and the model is built. The corresponding reconstruction error is calculated for each test sample by this model, and the abnormal behavior of examinee is detected by this model. In this paper, the experimental data set, this method can achieve a higher detection performance, and can achieve real-time detection speed. (4) aiming at the sparse reconstruction-based examinee abnormal behavior detection method performance deficiencies, In this paper, a motion connectivity region detection method based on motion history image is added to form a dual-channel detection framework based on multi-information fusion. Under the dual channel, the performance of examinee abnormal behavior detection is improved effectively. The proposed method is compared and analyzed with other common methods of examinee suspicious behavior detection. The experiments show that the proposed method has better universality.
【學(xué)位授予單位】：中國科學(xué)技術(shù)大學(xué)
【學(xué)位級別】：碩士
【學(xué)位授予年份】：2017
【分類號】：TP391.41;TN948.6

【相似文獻】

相關(guān)期刊論文 前10條

1 定志鋒;戴聲奎;陳美龍;;基于統(tǒng)計結(jié)構(gòu)梯度特征的行人檢測[J];通信技術(shù);2012年07期

2 鄧宇;李振波;李華;;圖切割支持的融合顏色和梯度特征的實時背景減除方法[J];計算機輔助設(shè)計與圖形學(xué)學(xué)報;2006年11期

3 吳亞順;王平;白朝元;孔融;;一種基于梯度特征的陰影消除方法[J];電腦編程技巧與維護;2014年17期

4 陳愛華;朱明;王艷華;薛陳;;融合梯度特征的灰度目標跟蹤[J];微電子學(xué)與計算機;2009年02期

5 李琨;鄭慶暉;廖冬學(xué);;基于梯度特征的圖像自動分割方法[J];宇航學(xué)報;2006年06期

6 余博;郭雷;趙天云;錢曉亮;左蔚;;融合梯度特征的紅外目標跟蹤[J];西安工業(yè)大學(xué)學(xué)報;2012年05期

7 郭偉;朱紅;趙亦工;李欣;謝振華;;基于新的梯度特征相似度量的目標跟蹤算法[J];光電工程;2008年04期

8 萬明;張鳳鳴;胡雙;;基于多步長梯度特征的紅外弱小目標檢測算法[J];光學(xué)學(xué)報;2011年10期

9 劉宏;李錦濤;劉群;錢躍良;李豪杰;;融合顏色和梯度特征的運動陰影消除方法[J];計算機輔助設(shè)計與圖形學(xué)學(xué)報;2007年10期

10 鄭茜穎;王欽敏;毛政元;;基于梯度特征的高分辨率衛(wèi)星遙感影像陰影檢測[J];福州大學(xué)學(xué)報(自然科學(xué)版);2008年05期

相關(guān)碩士學(xué)位論文 前4條

1 丁苗苗;基于考場監(jiān)控視頻的智能監(jiān)考方法研究[D];中國科學(xué)技術(shù)大學(xué);2017年

2 定志鋒;基于統(tǒng)計結(jié)構(gòu)梯度特征的行人檢測[D];華僑大學(xué);2012年

3 張竹偉;基于梯度特征和位置信息的目標識別算法研究[D];中北大學(xué);2015年

4 肖永剛;基于梯度特征和級聯(lián)分類的快速行人檢測[D];天津大學(xué);2010年

，

本文編號：2441962