【摘要】：人類(lèi)的偉大理想之一就是讓機(jī)器人可以具備像他們自己一樣的視覺(jué)功能。近一個(gè)世紀(jì)以來(lái),信息技術(shù)飛速發(fā)展,計(jì)算機(jī)視覺(jué)方面更是科研工作者們研究的重點(diǎn)。到今天,計(jì)算機(jī)視覺(jué)領(lǐng)域的目標(biāo)跟蹤技術(shù)在計(jì)算精確度和跟蹤實(shí)時(shí)性等方面已經(jīng)達(dá)到了較高的水平。目標(biāo)跟蹤的目的,就是針對(duì)長(zhǎng)度為不同時(shí)間的某些幀連續(xù)序列圖像,這些序列所包含的每幅圖像中均由需要被定位的運(yùn)動(dòng)目標(biāo)。對(duì)科研工作者來(lái)說(shuō)。只有符合以下標(biāo)準(zhǔn),視頻目標(biāo)跟蹤方法才能達(dá)標(biāo):1.實(shí)時(shí)性好,即其處理速度需要達(dá)到一定的數(shù)值;2.魯棒性強(qiáng),即面對(duì)復(fù)雜場(chǎng)景或目標(biāo)的姿勢(shì)、動(dòng)作均發(fā)生大幅度改變時(shí)也不會(huì)影響算法的穩(wěn)定性,仍然可以跟蹤到目標(biāo)。但與理論研究不同,在實(shí)際應(yīng)用中,視頻目標(biāo)跟蹤技術(shù)仍面對(duì)著諸如場(chǎng)景的復(fù)雜化,目標(biāo)的突然變化,光照變化等多個(gè)難題。本論文針對(duì)類(lèi)似對(duì)象干擾、動(dòng)態(tài)模糊、低對(duì)比度、部分遮擋和光照變化等實(shí)際生活中出現(xiàn)在被跟蹤目標(biāo)所在視頻序列的常見(jiàn)情況做出了分析研究,取得了一些主要研究成果,說(shuō)明如下:1.提出一種基于空間鄰域約束編碼的視頻跟蹤方法。該方法采用了新的約束策略,即通過(guò)加權(quán)碼進(jìn)行雙重加權(quán)的空間鄰域約束編碼模型,該模型是通過(guò)分別考慮特征像素的相鄰像素的灰度加權(quán)編碼及他們之間的歐式距離加權(quán)編碼來(lái)得到的。該模型除了考慮像素本身顏色值以外,還將距離這類(lèi)空間信息考慮在內(nèi),以獲得在復(fù)雜場(chǎng)景的幀提取對(duì)應(yīng)像素各種特征的健壯的代碼。它進(jìn)一步增強(qiáng)了編碼的穩(wěn)定性,使目標(biāo)跟蹤所使用的跟蹤器更加健壯,在進(jìn)行目標(biāo)跟蹤時(shí)取得了更加精確可靠的跟蹤效果。2.在空間鄰域約束編碼的基礎(chǔ)上,提出了其與Mean shift(均值漂移)綜合后跟蹤這樣一種視頻追蹤方式。本方法在利用空間鄰域約束編碼模型來(lái)得到目標(biāo)像素準(zhǔn)確編碼的同時(shí),加入了Mean shift算法。Mean shift算法擁有的優(yōu)勢(shì)為:1.運(yùn)算成本低,當(dāng)待追蹤標(biāo)的范圍確定時(shí),能夠以24幀/秒的速率進(jìn)行追蹤;2.即使目標(biāo)產(chǎn)生形變,角度偏移,其邊際不完全顯示等情況,該算法也會(huì)排除干擾,準(zhǔn)確追蹤這兩類(lèi)優(yōu)勢(shì)。從而在確保算法健壯性的基礎(chǔ)上提高了算法的實(shí)時(shí)性,使其面對(duì)復(fù)雜的場(chǎng)景時(shí)也可以做到準(zhǔn)確定位跟蹤。
[Abstract]:One of the great ideals of mankind is that robots can have the same visual function as they do. In the past century, with the rapid development of information technology, computer vision is the focus of researchers. Today, the target tracking technology in the field of computer vision has reached a high level in computing accuracy and real-time tracking. The purpose of target tracking is to target some successive sequence images of frames with different length of time. Each image contained in these sequences is made up of moving targets that need to be located. For researchers. Video target tracking can only meet the following standards: 1. Good real-time, that is, the processing speed needs to reach a certain value; 2. Robustness is strong, that is, facing the posture of complex scene or target, the stability of the algorithm will not be affected when the action changes greatly, and the target can still be tracked. However, different from the theoretical research, video target tracking technology still faces many difficulties in practical applications, such as the complexity of the scene, the sudden change of the target, the change of illumination, and so on. In this paper, we analyze and study the common situations of similar object interference, dynamic blur, low contrast, partial occlusion and illumination change, which appear in the video sequence of the target being tracked, and obtain some main research results. The description is as follows: 1. A video tracking method based on spatial neighborhood constrained coding is proposed. In this method, a new constraint strategy is adopted, that is, the spatial neighborhood constraint coding model is double weighted by weighted code. The model is obtained by taking into account the grayscale weighted coding of adjacent pixels and the Euclidean distance weighted coding between them respectively. In addition to considering the color value of pixels, the model also takes the spatial information of distance into account to obtain robust codes for extracting various features of pixels in the frames of complex scenes. It further enhances the stability of coding, makes the tracker used in target tracking more robust, and achieves a more accurate and reliable tracking effect. 2. Based on the spatial neighborhood constrained coding, this paper proposes a video tracking method which is integrated with Mean shift (mean shift. In this method, the spatial neighborhood constraint coding model is used to get the accurate coding of the target pixels, and the advantages of the Mean shift algorithm. Mean shift algorithm are as follows: 1. The operation cost is low, when the range of the target to be tracked is determined, it can be traced at the rate of 24 frames / second; 2. Even if the target produces deformation, angle deviation and incomplete display of the edge, the algorithm can eliminate the interference and track the two kinds of advantages accurately. In order to ensure the robustness of the algorithm on the basis of improving the real-time algorithm, so that it can also be accurate location tracking in the face of complex scenes.
【學(xué)位授予單位】：西安電子科技大學(xué)
【學(xué)位級(jí)別】：碩士
【學(xué)位授予年份】：2014
【分類(lèi)號(hào)】：TN919.81

【相似文獻(xiàn)】

相關(guān)期刊論文 前9條

1 李海洋;何東健;;一種自適應(yīng)空間鄰域的顯著圖獲取方法研究[J];計(jì)算機(jī)應(yīng)用研究;2013年06期

2 鄭孝俊;趙海峰;羅斌;;一種醫(yī)學(xué)三維數(shù)據(jù)集中感興趣空間鄰域體快速分割方法[J];計(jì)算機(jī)應(yīng)用與軟件;2011年09期

3 印勇;王亞飛;;基于空間鄰域相關(guān)性的運(yùn)動(dòng)目標(biāo)檢測(cè)方法[J];光電工程;2009年02期

4 孟麗敏;宋余慶;朱峰;;基于空間鄰域加權(quán)的模糊C-均值聚類(lèi)及其應(yīng)用研究[J];計(jì)算機(jī)應(yīng)用研究;2010年10期

5 蘇錦旗;薛惠鋒;吳慧欣;;基于熵度量的空間鄰域離群點(diǎn)查找[J];計(jì)算機(jī)工程與應(yīng)用;2009年21期

6 諶德榮;孫波;陶鵬;宮久路;;基于核光譜角余弦的高光譜圖像空間鄰域聚類(lèi)方法[J];電子學(xué)報(bào);2008年10期

7 曾小明,譚楓;馬爾可夫平穩(wěn)隨機(jī)域模型與遙感圖象空間鄰域結(jié)構(gòu)信息分析[J];信息與控制;1988年04期

8 楊?lèi)?郭樹(shù)旭;任瑞治;于永力;;基于核函數(shù)及空間鄰域信息的FCM圖像分割新算法[J];吉林大學(xué)學(xué)報(bào)(工學(xué)版);2011年S2期

9 ;[J];;年期

相關(guān)碩士學(xué)位論文 前1條

1 趙凡迪;基于空間鄰域約束編碼的視頻目標(biāo)跟蹤研究[D];西安電子科技大學(xué);2014年

，

本文編號(hào)：2389230