本文選題：多攝像機(jī)跟蹤　切入點(diǎn)：目標(biāo)交接　出處：《聊城大學(xué)》2015年碩士論文　論文類型：學(xué)位論文

【摘要】：基于視頻監(jiān)控的運(yùn)動(dòng)目標(biāo)跟蹤技術(shù)屬于人工智能的分支,涉及到模式識(shí)別、圖像處理、計(jì)算機(jī)視覺、機(jī)器學(xué)習(xí)等學(xué)科。目前,單攝像機(jī)的目標(biāo)檢測(cè)和跟蹤技術(shù)發(fā)展的比較成熟,取得了大量的研究成果。但是,仍有一系列的難題需要解決。比如,目標(biāo)之間出現(xiàn)遮擋、復(fù)雜背景的干擾、動(dòng)態(tài)背景情況的跟蹤等。多攝像機(jī)目標(biāo)跟蹤是近年來(lái)的研究熱點(diǎn),而目標(biāo)交接是實(shí)現(xiàn)目標(biāo)匹配和場(chǎng)景切換跟蹤的關(guān)鍵技術(shù),是多攝像機(jī)監(jiān)控必須解決的難題。本課題在研究單攝像機(jī)運(yùn)動(dòng)目標(biāo)檢測(cè)、跟蹤算法基礎(chǔ)上提出了新的方法,并重點(diǎn)研究了雙攝像視野重疊情況下的視野分界線生成和目標(biāo)交接算法,主要工作如下:1.采用了融合的目標(biāo)檢測(cè)算法。首先采用幀間差分判斷出目標(biāo)所在位置的大致范圍,并鎖定中心位置,然后在此范圍附近,采用背景差分的方法,準(zhǔn)確判斷目標(biāo)區(qū)域。采用此種方法的優(yōu)點(diǎn)是,對(duì)目標(biāo)區(qū)域的判斷準(zhǔn)確,且受外界擾動(dòng)的影響較小,可以得到較為完整的目標(biāo)前景區(qū)域。2.采用了粒子濾波改進(jìn)的Camshift目標(biāo)跟蹤算法,結(jié)合Camshift算法對(duì)粒子濾波算法進(jìn)行改進(jìn)。同時(shí)解決了粒子濾波需要計(jì)算大量粒子,粒子收斂速度慢,跟蹤實(shí)時(shí)性差問題和遮擋情況下Camshift算法目標(biāo)跟蹤魯棒性低問題。3.提出了基于SURF快速特征點(diǎn)匹配和投影不變量的視野分界線生成算法。首先利用SURF算法快速獲得相鄰兩個(gè)攝像機(jī)中帶有重疊區(qū)域的背景圖像的特征點(diǎn),然后利用RANSAC算法對(duì)特征點(diǎn)向量進(jìn)行優(yōu)化,去除匹配誤差較大的點(diǎn)。通過對(duì)RANSAC算法的迭代,選擇出4對(duì)最佳匹配點(diǎn),最后,根據(jù)兩幅背景圖像的邊界點(diǎn)以及投影不變量求兩幅圖像中重疊區(qū)域,并進(jìn)行標(biāo)定,生成視野分界線。4.提出了基于視野分界線幾何坐標(biāo)變換曲線匹配方法的目標(biāo)交接算法。首先實(shí)時(shí)檢測(cè)公共視野區(qū)域內(nèi)是否有運(yùn)動(dòng)目標(biāo)出現(xiàn),然后進(jìn)行坐標(biāo)變換,并在新的坐標(biāo)系下,對(duì)目標(biāo)像素進(jìn)行統(tǒng)計(jì),最后,將得到的像素統(tǒng)計(jì)曲線進(jìn)行匹配和標(biāo)定。
[Abstract]:The branch target video surveillance tracking technology is based on artificial intelligence, involves pattern recognition, image processing, computer vision, machine learning and other disciplines. At present, the target detection and tracking technology development of single camera is relatively mature, made a lot of research results. However, there are still a series of problems need to be solved. For example between, target occlusion, complex background and dynamic background tracking. Multi camera tracking is a hot research topic in recent years, but the goal is to achieve the goal of key technology transfer, and scene change tracking, multi camera surveillance is a difficult problem to be solved. This topic in the study of single camera moving target detection, tracking the algorithm is proposed based on the new method, and focuses on the dual camera view overlap under the vision line generation and target handoff algorithm, the main work is as follows 1.: the target detection algorithm fusion. Firstly, using frame difference estimate approximate range of target position, and the locking range near the center position, and then, using the background difference method, accurately determine the target area. The advantages of this method is that the target area is accurate, and less influence by the external disturbance, the target tracking algorithm can get more complete foreground.2. using the improved particle filter Camshift, combined with Camshift algorithm to improve particle filter algorithm. At the same time to solve the particle filter need to calculate a large number of particles, the particles slow convergence of Camshift algorithm and target occlusion tracking real-time robust tracking problems the problem of low.3. proposed SURF fast feature point matching and projective invariants based on the view of demarcation line generation algorithm. SURF algorithm is first used to obtain fast The background image feature points with overlap area of two adjacent the camera, and then use RANSAC algorithm to optimize feature vector, remove the matching error larger. Through the iteration of the RANSAC algorithm, selected 4 of the best matching point, finally, according to the overlap area of the two painting like boundary points and projective invariants for the two images, and calibrate the generated vision line.4. proposed target handoff algorithm vision line geometric coordinate transformation curve matching method based on real-time detection. The first public view area if there is a moving target, and then the coordinate transformation, and in the new coordinate system, statistics, on target pixel finally, the statistics of pixel curve obtained by matching and calibration.

【學(xué)位授予單位】：聊城大學(xué)
【學(xué)位級(jí)別】：碩士
【學(xué)位授予年份】：2015
【分類號(hào)】：TP391.41;TN948.6

【參考文獻(xiàn)】

相關(guān)期刊論文 前3條

1 王兆光;王敬東;李鵬;;一種Camshift優(yōu)化的粒子濾波跟蹤算法[J];光電子技術(shù);2010年01期

2 付輝敬;田錚;冉茂華;康召輝;;閉塞目標(biāo)識(shí)別中的仿射不變曲線匹配[J];光電工程;2013年03期

3 初紅霞;謝忠玉;王希鳳;李占英;李欣;;基于改進(jìn)粒子濾波的多目標(biāo)跟蹤算法研究[J];計(jì)算機(jī)工程與設(shè)計(jì);2014年06期

，

本文編號(hào)：1615172