【摘要】：本文使用粒子熵方法來描述群體的分布信息,該方法首先將粒子均勻地灑在每一幀視頻上,用視頻幀中粒子的流動(dòng)來描述視頻中群體的運(yùn)動(dòng),粒子速度的計(jì)算是根據(jù)周圍像素點(diǎn)的光流值計(jì)算得到的,文中把運(yùn)動(dòng)速度大于某個(gè)閾值的粒子定義為運(yùn)動(dòng)粒子。將運(yùn)動(dòng)的粒子分別投影到水平與垂直坐標(biāo)軸上,計(jì)算粒子在水平與垂直方向上的概率分布,通過粒子的概率分布計(jì)算得到粒子的熵值,用運(yùn)動(dòng)粒子的熵值來描述人群的分布信息。最后,根據(jù)粒子的運(yùn)動(dòng)速度計(jì)算得到粒子的運(yùn)動(dòng)動(dòng)能。通常情況下正常人群行為的數(shù)量遠(yuǎn)遠(yuǎn)大于異常人群行為的數(shù)量,因而群體異常行為檢測是一個(gè)不平衡問題。根據(jù)高斯混合模型(GMM)在處理不平衡問題上的優(yōu)勢,本文利用GMM對正常群體行為進(jìn)行建模。在建模階段使用的訓(xùn)練樣本中只包含正常的群體行為,利用人群分布的粒子熵值與人群運(yùn)動(dòng)的動(dòng)能來分別建立正常群體行為的高斯混合模型。在異常群體行為檢測階段,提取出待檢測視頻的人群分布特征值與人群運(yùn)動(dòng)動(dòng)能特征值,使用提取的特征值在建模階段建立的高斯混合模型上計(jì)算其概率,若在兩個(gè)模型上計(jì)算得到的概率值都小于閾值時(shí),則該特征所對應(yīng)的視頻序列中有異常群體行為。本文在包含聚集與分散事件的公共可用數(shù)據(jù)集UMN數(shù)據(jù)集與PETS2009數(shù)據(jù)集上進(jìn)行了異常群體行為檢測的實(shí)驗(yàn),實(shí)驗(yàn)驗(yàn)證了本文方法能有效、準(zhǔn)確地檢測出群體的異常行為。
[Abstract]:In this paper, the particle entropy method is used to describe the distribution information of the group. Firstly, the particles are sprinkled evenly on each frame of video, and the flow of particles in the video frame is used to describe the motion of the group in the video. The calculation of particle velocity is based on the optical flow value of surrounding pixel points. In this paper, the particle whose velocity is greater than a certain threshold is defined as moving particle. The moving particles are projected on the horizontal and vertical coordinate axis respectively, the probability distribution of the particles in the horizontal and vertical directions is calculated, and the entropy value of the particles is obtained by calculating the probability distribution of the particles. The entropy value of moving particles is used to describe the distribution information of the population. Finally, the kinetic energy of the particle is calculated according to the velocity of the particle. In general, the number of normal population behavior is much larger than that of abnormal population behavior, so group abnormal behavior detection is an unbalanced problem. According to the advantages of Gao Si mixed model (GMM) in dealing with unbalanced problems, this paper uses GMM to model the behavior of normal groups. The training samples used in the modeling stage only contain the normal group behavior. The Gao Si mixed model of the normal group behavior is established by using the particle entropy value of the population distribution and the kinetic energy of the population movement respectively. In the stage of abnormal group behavior detection, the eigenvalues of population distribution and kinetic energy of the video to be detected are extracted, and the probability of the extracted eigenvalues is calculated on the Gao Si hybrid model established in the modeling stage. If the probability values calculated on both models are less than the threshold, then the video sequence corresponding to this feature has abnormal group behavior. In this paper, the experiments of abnormal group behavior detection are carried out on the common available dataset UMN dataset and PETS2009 dataset containing aggregation and decentralized events. The experimental results show that the proposed method can detect the abnormal behavior of the group effectively and accurately.
【學(xué)位授予單位】：西安理工大學(xué)
【學(xué)位級別】：碩士
【學(xué)位授予年份】：2016
【分類號】：TP391.41

【參考文獻(xiàn)】

相關(guān)期刊論文 前1條

1 周潔;郭立君;張榮;;基于粒子熵值的異常行為檢測[J];無線電通信技術(shù);2015年03期

相關(guān)博士學(xué)位論文 前1條

1 葉子;基于視覺計(jì)算的人群場景理解與行為分析[D];中國礦業(yè)大學(xué)（北京）;2013年

相關(guān)碩士學(xué)位論文 前1條

1 古緒新;基于人群速度與人群分布的異常人群行為檢測[D];哈爾濱工業(yè)大學(xué);2013年

，

本文編號：2484720