本文選題：目標(biāo)檢測(cè) + 深度學(xué)習(xí)��； 參考：《北京交通大學(xué)》2017年碩士論文

【摘要】：目標(biāo)檢測(cè)作為圖像處理和計(jì)算機(jī)視覺領(lǐng)域中的經(jīng)典課題,在交通監(jiān)控、圖像檢索、人機(jī)交互等方面有著廣泛的應(yīng)用。它旨在一個(gè)靜態(tài)圖像(或動(dòng)態(tài)視頻)中檢測(cè)出人們感興趣的目標(biāo)對(duì)象。傳統(tǒng)的目標(biāo)檢測(cè)算法中特征提取和分類決策分開進(jìn)行,對(duì)特征選取的要求就更加嚴(yán)格,在面對(duì)復(fù)雜場(chǎng)景的時(shí)候很難得到理想效果。自Hinton教授提出深度學(xué)習(xí)理論,越來越多的研究學(xué)者嘗試采用深度學(xué)習(xí)理念來解決目標(biāo)檢測(cè)問題,并且提出了不同的模型。不同的模型應(yīng)用也不盡相同,通常采用卷積神經(jīng)網(wǎng)絡(luò)來處理目標(biāo)檢測(cè)問題。相比于傳統(tǒng)目標(biāo)檢測(cè)算法,卷積神經(jīng)網(wǎng)絡(luò)中特征提取和模式分類并行進(jìn)行,而且隨著層數(shù)的增多可以更好的處理復(fù)雜場(chǎng)景,但是它對(duì)目標(biāo)邊緣的約束性太差。在這樣的基礎(chǔ)上,本文對(duì)傳統(tǒng)算法和卷積神經(jīng)網(wǎng)絡(luò)做了深入的研究,實(shí)現(xiàn)了將傳統(tǒng)算法和卷積神經(jīng)網(wǎng)絡(luò)相結(jié)合的目標(biāo)檢測(cè)算法。本文的主要工作和創(chuàng)新有:(1)針對(duì)傳統(tǒng)目標(biāo)檢測(cè)算法一般是使用矩形框的方式得到目標(biāo)的大致區(qū)域,而我們的需求是盡可能的獲得目標(biāo)的邊緣輪廓問題,本文實(shí)現(xiàn)了一種改進(jìn)的基于主動(dòng)輪廓模型的目標(biāo)檢測(cè)算法,使輪廓盡可能的接近目標(biāo)。(2)針對(duì)傳統(tǒng)算法需要人工設(shè)計(jì)圖像特征,不同場(chǎng)景模型不穩(wěn)定問題而卷積神經(jīng)網(wǎng)絡(luò)分割不精確以及缺少相鄰像素之間的約束的問題,本文將傳統(tǒng)算法和卷積神經(jīng)網(wǎng)絡(luò)相結(jié)合,使用卷積神經(jīng)網(wǎng)絡(luò)進(jìn)行圖像“高層次”特征提取,超像素提取出圖像“低層次”特征,可以適應(yīng)不同的復(fù)雜場(chǎng)景,并且獲得準(zhǔn)確的目標(biāo)邊緣。在頤和園景點(diǎn)數(shù)據(jù)庫(kù)中進(jìn)行了充分的實(shí)驗(yàn)。通過結(jié)果可以看出使用我們的算法進(jìn)行目標(biāo)檢測(cè)提取,可以很精確的提取出目標(biāo),而且目標(biāo)的邊緣約束性也非常強(qiáng)。(3)主要?jiǎng)?chuàng)新點(diǎn):在使用超像素提取特征刪除重復(fù)特征,減少特征冗余,降低特征的維度;同時(shí)將VGGNet(Visual Geometry Group Network)網(wǎng)絡(luò)換成收斂速度更快的GoogleNet網(wǎng)絡(luò),提高了算法的速度。
[Abstract]:As a classical subject in the field of image processing and computer vision, target detection has been widely used in traffic monitoring, image retrieval, human-computer interaction and so on. It aims to detect the object of interest in a static image (or dynamic video). In the traditional target detection algorithm, the feature extraction and classification decision are carried out separately, the requirement of feature selection is more strict, and it is difficult to obtain ideal results in the face of complex scene. Since Professor Hinton put forward the theory of deep learning, more and more researchers have tried to solve the problem of target detection by using the concept of deep learning, and put forward different models. The application of different models is different. Convolution neural network is usually used to deal with the target detection problem. Compared with the traditional target detection algorithm, the convolution neural network features extraction and pattern classification parallel, and with the increase of the number of layers can better deal with the complex scene, but it is too bad to target edge constraint. On this basis, the traditional algorithm and the convolutional neural network are studied in this paper, and the target detection algorithm combining the traditional algorithm and the convolution neural network is realized. The main work and innovation of this paper are: 1) for traditional target detection algorithms, we usually use rectangular frame to get the approximate area of the target, and our requirement is to obtain the edge contour of the target as much as possible. In this paper, an improved target detection algorithm based on active contour model is implemented, which makes the contour as close as possible to the target. In this paper, the traditional algorithm and convolution neural network are combined to solve the problems of unstable scene models and inaccurate segmentation of convolutional neural networks and the lack of constraints between adjacent pixels. Using convolution neural network to extract "high level" feature of image, super-pixel can extract "low level" feature of image, which can adapt to different complex scene and obtain accurate target edge. A full experiment has been carried out in the Summer Palace scenic spot database. The results show that using our algorithm for target detection and extraction, we can extract the target accurately, and the edge constraint of the target is also very strong. At the same time, the VGGNet(Visual Geometry Group Network) network is replaced by a faster convergent GoogleNet network, which improves the speed of the algorithm.
【學(xué)位授予單位】：北京交通大學(xué)
【學(xué)位級(jí)別】：碩士
【學(xué)位授予年份】：2017
【分類號(hào)】：TP391.41;TP183

【參考文獻(xiàn)】

相關(guān)期刊論文 前5條

1 常戩;白佳弘;;基于回轉(zhuǎn)對(duì)稱雙邊濾波的Retinex圖像增強(qiáng)算法[J];計(jì)算機(jī)工程;2016年06期

2 姜建國(guó);趙媛;孟宏偉;李博;;采用類電磁機(jī)制算法的SVM決策樹多分類策略[J];西安電子科技大學(xué)學(xué)報(bào);2014年06期

3 李亞珂;趙海武;李國(guó)平;滕國(guó)偉;王國(guó)中;;一種有效的自適應(yīng)均值濾波算法[J];電視技術(shù);2013年23期

4 付忠良;;不平衡多分類問題的連續(xù)AdaBoost算法研究[J];計(jì)算機(jī)研究與發(fā)展;2011年12期

5 曾春;李曉華;周激流;;基于感興趣區(qū)梯度方向直方圖的行人檢測(cè)[J];計(jì)算機(jī)工程;2009年24期

，

本文編號(hào)：1932251