基于核化相關(guān)濾波器的目標(biāo)跟蹤技術(shù)研究及應(yīng)用系統(tǒng)
發(fā)布時間：2019-02-14 16:38
【摘要】：基于視頻的目標(biāo)跟蹤技術(shù)是指使用計算機對視頻序列中感興趣目標(biāo)進行持續(xù)跟蹤從而得到目標(biāo)運動軌跡等信息。目標(biāo)跟蹤技術(shù)在安防視頻監(jiān)控、視頻文件壓縮、人機交互、智能交通等領(lǐng)域都有廣泛的應(yīng)用。然而,由于視頻背景環(huán)境的復(fù)雜性和目標(biāo)在運動過程中經(jīng)常會改變外觀和形狀等因素的影響,導(dǎo)致對目標(biāo)實現(xiàn)準確的跟蹤是一件非常困難的事情。本文主要對運動目標(biāo)檢測與跟蹤算法在目標(biāo)尺度變化、目標(biāo)受到其它物體遮擋等場景下存在的一些問題進行研究。本文的主要研究工作內(nèi)容如下:1、單目標(biāo)跟蹤方面。本文首先對核化相關(guān)濾波器(Kernelized Correlation Filter,KCF)跟蹤算法進行了深入的研究并提出一種遮擋檢測算法判定所跟蹤目標(biāo)是否受到其它物體的遮擋干擾,進而使用卡爾曼濾波器對受到遮擋干擾的目標(biāo)進行位置預(yù)測。通過實驗可知,加入遮擋檢測后的核化相關(guān)濾波器跟蹤算法較好地解決了目標(biāo)在線性運動下受到其它物體遮擋干擾的問題;然后,本文使用相關(guān)濾波器(CorrelationFilter,CF)和圖像尺度金字塔計算目標(biāo)的尺度大小,較好地解決了目標(biāo)在運動過程中存在的尺度變化問題。2、多目標(biāo)跟蹤方面。本文首先在運動目標(biāo)檢測算法中融入深度圖像,較好解決了運動目標(biāo)的陰影干擾。接著,本文提出了一種基于核化相關(guān)濾波器(KCF)跟蹤算法和運動目標(biāo)檢測的多目標(biāo)跟蹤算法。在目標(biāo)之間沒有相互遮擋情況下,使用運動目標(biāo)檢測算法得到運動目標(biāo)區(qū)域,在運動目標(biāo)區(qū)域上進行人臉檢測后使用卡爾曼濾波器和數(shù)據(jù)關(guān)聯(lián)技術(shù)生成多目標(biāo)軌跡;一旦判定多個目標(biāo)將要交互,對每個目標(biāo)都使用KCF跟蹤算法進行獨立跟蹤。實驗結(jié)果表明,提出的多目標(biāo)跟蹤算法能夠較好地處理多目標(biāo)之間存在的遮擋問題。3、最后本文設(shè)計一個基于核化相關(guān)濾波器目標(biāo)跟蹤算法的應(yīng)用系統(tǒng),并實現(xiàn)了基于視頻目標(biāo)跟蹤技術(shù)的視頻監(jiān)控網(wǎng)絡(luò)微信報警子系統(tǒng)、跑步視頻分析子系統(tǒng)和多目標(biāo)人數(shù)統(tǒng)計子系統(tǒng)。驗證了本文所提出的改進目標(biāo)跟蹤算法在現(xiàn)實生活場景中的可行性。
[Abstract]:Video based target tracking technology refers to the continuous tracking of objects of interest in video sequences using computers to obtain information such as the moving trajectory of the target. Target tracking technology has been widely used in security video surveillance, video file compression, human-computer interaction, intelligent transportation and other fields. However, due to the complexity of the video background environment and the influence of factors such as changing the appearance and shape of the target in the process of moving, it is very difficult to track the target accurately. In this paper, some problems in moving target detection and tracking algorithms are studied, such as the change of the target scale and the occlusion of the target by other objects. The main work of this paper is as follows: 1. Single target tracking. In this paper, we study the (Kernelized Correlation Filter,KCF (Kernelized correlation filter) tracking algorithm and propose an occlusion detection algorithm to determine whether the target is affected by other objects. Furthermore, Kalman filter is used to predict the position of the target affected by occlusion interference. The experimental results show that the kernel correlation filter tracking algorithm with occlusion detection can solve the problem that the target is blocked by other objects under linear motion. Then, the correlation filter (CorrelationFilter,CF) and the image scale pyramid are used to calculate the size of the target, which can solve the problem of the scale change in the moving process. 2. Multi-target tracking. Firstly, the depth image is incorporated into the moving target detection algorithm, which solves the shadow interference of moving target. Then, a multi-target tracking algorithm based on Kernel correlation filter (KCF) and moving target detection is proposed. When there is no mutual occlusion between targets, moving target detection algorithm is used to obtain moving target region, and Kalman filter and data association technique are used to generate multi-target trajectory after face detection in moving target area. Once it is determined that multiple targets will interact, each target is tracked independently using the KCF tracking algorithm. Experimental results show that the proposed multi-target tracking algorithm can deal with the occlusion problem between multiple targets. 3. Finally, this paper designs an application system based on the Kernel correlation filter target tracking algorithm. The video surveillance network WeChat alarm subsystem, running video analysis subsystem and multi-target statistics subsystem are implemented based on video target tracking technology. The feasibility of the proposed improved target tracking algorithm in real life scenarios is verified.
【學(xué)位授予單位】：北京郵電大學(xué)
【學(xué)位級別】：碩士
【學(xué)位授予年份】：2016
【分類號】：TP391.41
[Abstract]:Video based target tracking technology refers to the continuous tracking of objects of interest in video sequences using computers to obtain information such as the moving trajectory of the target. Target tracking technology has been widely used in security video surveillance, video file compression, human-computer interaction, intelligent transportation and other fields. However, due to the complexity of the video background environment and the influence of factors such as changing the appearance and shape of the target in the process of moving, it is very difficult to track the target accurately. In this paper, some problems in moving target detection and tracking algorithms are studied, such as the change of the target scale and the occlusion of the target by other objects. The main work of this paper is as follows: 1. Single target tracking. In this paper, we study the (Kernelized Correlation Filter,KCF (Kernelized correlation filter) tracking algorithm and propose an occlusion detection algorithm to determine whether the target is affected by other objects. Furthermore, Kalman filter is used to predict the position of the target affected by occlusion interference. The experimental results show that the kernel correlation filter tracking algorithm with occlusion detection can solve the problem that the target is blocked by other objects under linear motion. Then, the correlation filter (CorrelationFilter,CF) and the image scale pyramid are used to calculate the size of the target, which can solve the problem of the scale change in the moving process. 2. Multi-target tracking. Firstly, the depth image is incorporated into the moving target detection algorithm, which solves the shadow interference of moving target. Then, a multi-target tracking algorithm based on Kernel correlation filter (KCF) and moving target detection is proposed. When there is no mutual occlusion between targets, moving target detection algorithm is used to obtain moving target region, and Kalman filter and data association technique are used to generate multi-target trajectory after face detection in moving target area. Once it is determined that multiple targets will interact, each target is tracked independently using the KCF tracking algorithm. Experimental results show that the proposed multi-target tracking algorithm can deal with the occlusion problem between multiple targets. 3. Finally, this paper designs an application system based on the Kernel correlation filter target tracking algorithm. The video surveillance network WeChat alarm subsystem, running video analysis subsystem and multi-target statistics subsystem are implemented based on video target tracking technology. The feasibility of the proposed improved target tracking algorithm in real life scenarios is verified.
【學(xué)位授予單位】：北京郵電大學(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】
相關(guān)期刊論文 前2條
1 蔣戀華;甘朝暉;蔣e，
本文編號：2422395
本文鏈接：http://sikaile.net/kejilunwen/ruanjiangongchenglunwen/2422395.html
最近更新
教材專著
