［6］ 代科学，李国辉，涂丹． 监控视频运动目标检测减背景技术的研究现状和展望［J］． 中国图象图形学报，2006，11( 7) : 919-917．
［7］ ZHANG Hai-qing，LI Hou-qiang． Target tracking based on MonteCarlo method［J］． China Journal of Image and Graphics，2008，13( 5) : 937-938． ［8］ CASTLEMAN K R． 数字图像处理［M］． 朱志刚，林学闫，石定机，等译． 北京: 电子工业出版社，2002: 181-190． ［9］ HAN B，COMANICIU D，ZHU Ying，et al． Sequential kernel densityapproximation and its application to real-time visual tracking［J］．IEEE Trans on Pattern Analysis and Machine Intelligence，2008，30( 7) : 1186-1197．
［10］杨学超，刘文萍． 视频图像序列中的运动目标检测技术［J］． 计算机应用与软件，2008，25( 1) : 215-21
[11] 高素萍 智能小区安防系统的设计与实现[J] 电器应用 2006 [12] 章毓晋 图像处理和分析基础[M] 北京高等教育出版社 2002.7 [13] 杨杰 数字图像处理及matlab实现[M] 电子工业出版社 2011-1
[14]Lipton A,Fujiyoshi H and Patil R.Moving target classification and tracking from real-time video.In:Proc IEEE Workshop on Application of Computer Vision,Princeton,NJ,1998,8-14.
[15]任明武，孙涵。复杂背景下一种实用的运动目标检测方法。 计算机工程。2005.10 [16]Collins Retal.A system for video surveillance and monitoring:VSAM final report.Camegie Mellon University,Technical Report:CMU-RI=TR-00-12,2000. [16] 刘瑞祯/于仕琪 OpenCV教程 北京航空航天大学出版社 [M] 2007-6 [17] 刘翔 基于视频的运动目标检测与跟踪算法研究与应用[J] 2009-5 [18] 陈肃慧 视频目标检测与跟踪方法研究[J] 2011-12 [19] 湛为芳 C语言程序设计技术[M] 清华大学出版社 2006-3
[20] 赵俊 智能视频监控系统关键技术研究[D] 西安电子科技 大学 2007 [21] JAMESWC，JOHNWT．Analgorithmforthemachinecalculation ofcomplexFourierseries［J］．MathematicsofComputation，1965，19(2):297-301． [22]HPCchallengebenchma日［EB/OL］．［2010-11-01］．http://icl． cs．utk．edu/hpcc/． [23]NASparallelbenchmarks［EB/OL］．［2010-05-23］．http://www． nas．nasa．gov/Resources/Software/npb．html． [24]General-purposecomputationusinggraphicshardware［EB/OL］． ［2010-05-23］．http://www．gpgpu．org/． [25]KENNETHM，EDWARDA．TheFFTonaGPU［C］//Procofthe ACMSIGGRAPH/EUROGRAPHICSConferenceonGraphicsHard-ware．SanDiego，California:EurographicsAssociation，2003:112-119． [26]CUDAprogrammingguideversion3．0［K］．［S．l．］:NVIDIACorp， 2009． [27]CUDACUFFTLibrary［R］．［S．l．］:NVIDIACorp，2007． [28]FFTW［EB/OL］．［2010-05-23］．http://www．fftw．org/． [28]RAJESHB．Optimizingsparsematrix-vectormultipli-cationonGPUs［ R］．［S．l．］:IBM，2008
参考外文：
An Improved Moving Object Detection Algorithm Based on Frame Difference and Edge Detection
Moving object detection is very important in intelligent surveillance. In this paper, an improved algorithm based on frame difference and edge detection is presented for moving object detection. First of all, it detects the edges of each two continuous frames by Canny detector and gets the difference between the two edge images. And then, it divides the edge difference image into several small blocks and decides if they are moving areas by comparing the number of non-zero pixels to a threshold. At last, it does the block-connected component labeling to get the smallest rectangle that contains the moving object. Experimental results show the improved algorithm overcomes the shortcomings of the frame difference method. It has a high recognition rate and a high detection speed, which has a broad market prospect.
INTRODUCTION
Moving object detection from video sequences is an important research portion, since it can be used in many regions such as intelligent video surveillance, motion analysis,
human-machine interface applications, and so on. Moving object detection is the basis of moving object identification and tracking. Although a lot of studies have been conducted in recent years, the subject is still challenging [1], [2], [3].
Currently, the main detection algorithms include frame difference method, background subtraction method, optical flow method and statistical learning method. Optical flow method is the most complex algorithm. It spends more time than other methods, and statistical learning method needs many training samples and also has much computational complexity. These two methods are not suitable for real-time processing. Background subtraction method is extremely sensitive to the changes of light. Frame difference method is simple and easy to implement, but the results are not accurate enough, because the changes taking place in the background brightness cause misjudgment [4], [5], [6], [7]. According to that eyes are sensitive to both of movement and edges, this paper presented an improved algorithm that is combination of frame difference and edge detection. The algorithm has improved the object segmentation and object locating. It has a small calculation and a good detection speed, so it can do real-time processing. Because the edges of moving objects are almost not changed with light, this algorithm is not sensitive to the changes of light.
This paper firstly introduces two moving object detection algorithms of fixed scenes -- frame difference method and moving edge method and analyzes their advantages and disadvantages, and then presents a new algorithm based on them, lastly gives the experimental results and analysis.
TYPICAL MOVING OBJECT DETECTION ALGORITHMS
2.1 Frame difference method
To detect moving object in the surveillance video captured by immobile camera, the simplest method is the frame difference method for the reason that it has great detection speed, can be implemented on hardware easily and has been used widely[5].
While detecting moving object by frame difference method, in the difference image, the unchanged part is eliminated while the changed part remains. This change is caused by movement or noise, so it calls for a binary process upon the difference image to distinguish the moving objects and noise. Further more, connected component labeling is also needed to acquire the smallest rectangle containing the moving objects.
The noise is assumed as Gaussian white noise in calculating the threshold of the binary process. According to the theory of statistics, there is hardly any pixel which has dispersion more than 3 times of standard deviation. Thus the threshold is calculated as following:
TeX Source
While is the mean of the difference image, image.
is the standard deviation of the difference
The flow chart of the detecting process by frame difference method is as Figure 1.
Figure 1 The flow chart of frame difference method View All | Next
2.2 Moving edge method
Moving edge is the edge of moving objects. Difference image can be regarded as time gradient, while edge image is space gradient. Moving edge can be defined by the logic AND operation of difference image and the edge image [6].
The advantage of frame difference method is its small calculation, and the disadvantage is that it is sensitive to the noise. If the objects do not move but the brightness of the background changes, the results of frame difference methods may be not accurate enough. Since the edge has no relation with the brightness, moving edge method can overcome the disadvantage of frame difference method.
The flow chart of the detecting process by moving edge method is as Figure 2.
Figure 2 The flow chart of moving edge method Previous | View All | Next
To practical images, because of the existence of noise, normally the image filtering is needed. However, the filter will generate fuzzy edge. Traditional edge detectors such as Sobel, Prewitt, and Roberts do not give the accurate edges, but the edges which have a certain width. To be precise width of a single pixel edge, we can use Canny detector. Canny detector is one of the best edge detectors, which in many areas has been widely used in image processing [6]. The basic steps of detection by using Canny detector are as follows,
1. Filter the image by using Gaussian filter to remove the noise.
2. Calculate the ample and direction of the gradient of each pixel of the image after
filtering.
3. Make the “non-maximal inhibition” to the gradient in order to make the edge down to
one pixel width.
4. Do double threshold processing and edge linking.
IMPROVED MOVING OBJECT DETECTION ALGORITHM BASED ON FRAME DIFFERENCE AND EDGE DETECTION
Moving edge method can effectively suppress the noise caused by light, but it still has some misjudgments to some other noise. This paper proposes an improved algorithm based on frame difference and edge detection. Upon analysis, the method has better noise suppression and higher detection accuracy.
3.1 Algorithm introduction
The flow chart of the detection process by using the method based on frame difference and edge detection presented in this paper is as figure 3.
Figure 3 The flow chart of the improved algorithm Previous | View All | Next
The steps of new algorithm presented in this paper are as follows.
1. Get edge images and by edge detection with two continuous frames
and by using Canny edge detector. 2. Get edge difference image by difference between and . 3. Divide edge difference image into some certain small blocks and count the
number of non-zero pixels in the block, and recorded it as 4. If
.
5. Do connected components labeling to
that are too small.
6. Get the smallest rectangles containing the moving objects.
The algorithm has improved both the object segmentation and object locationing. Following is a brief analysis.
, and remove the connected components
,
is lager than the threshold, mark the block is a moving area, otherwise it is a
static area. Let 1 presents moving area and 0 presents static area, we can get a matrix
3.2 Object segmentation
Object segmentation is to divide the image into moving area and static area. The algorithm presented in this paper will get the edge images first, then difference them to get the edge difference image. In the final image we get, the pixel value of background area equal to 0 and pixel value of the edge of moving objects equal to 1. Now we will compare the difference between our algorithm and moving edge method mentioned in 2.2.
1. In moving edge method, assume two continuous frames are
background is
, moving objects are
and
and
,
, and independent white noise