《目标检测》作文（精选8篇）
My Favourite Teacher
I’ve got many good teachers.But my favourite one is our English teacher, Miss Lin.She’s about thirty years old and she looks very pretty.She is quite tall with short hair and a pair of glasses.She often wears jeans and T-shirt.She likes singing and doing sports.She works hard and she is very strict with us.She spends a lot of time taking care of us and helping us with our lessons.She often talks to us and encourages us when we get bad marks or do something wrong.Anyway, she is not only our good teacher but also our good friend.She is one of the best teachers in our school.I’m proud of her.Module 2
Dear Tony,How are you? I’m having a wonderful time in Beijing now.I have done many interesting things here.I have been to many places of interest.I have visited the Tian’anmen Square, the Palace Museum and the Great Wall.I have taken many nice photos of these famous places.I’ll send some to you with this email.I have also enjoyed different kinds of Chinese food.I’ve tried dumplings, noodles and roast duck.They are all very delicious.I think Beijing is one of the biggest and busiest cities in the world.There’s a lot to see and do here.I hope we can visit Beijing together next time.Yours,Tom
Module 3
Scientists think that there has been life on earth for hundreds of millions of years.Seven other planets go around the sun like the earth.However, none of them has an environment like that of the earth.So scientists do not think they will find life on them.The sun and its planets are called the solar system.And our solar system is a small part of a much larger group of stars and planets.We call it Galaxy.Scientists have also discovered many other galaxies in the universe.So it is impossible to imagine how large the universe is.We have always asked the question: With so many stars in the universe, are we alone? Is there life out there in the space? We’re looking forward to know the answer.Module 4
Health is very important for us all.In order to keep healthy, I always eat healthy food and do enough exercise.I have the three meals at home and I often eat vegetables and fruit, as well as meat and fish.After school, I often do some sports with my classmates, such as playing basketball and table tennis.In the evening, I go to bed early after I finish my homework.So I’m in excellent condition.Do you want to be as healthy as I? Here is some advice.First, stop eating fast food
and have healthy food.Second, don’t spend too much time in front of the computer.Try to do more exercise every day.Third, it’s good for your health to go to bed early in the evening.If you follow my advice, you will become healthier and healthier.Module 5
Milk and Coca are in the Sea
There are two main Characters, Milk and Coca, in my cartoon story.They are good friends.Milk is a lovely girl.She has a white face with two big eyes.She is not only kind, but also smart.She likes traveling in the sea.Coca is a cute boy.He has a black face with two small eyes.He is clever and helpful.He likes traveling in the sea, too.Both of them decide to have a trip in the sea, so they run away from their homes.In the sea they meet strange animals and sea plants.All of them are growing in the polluted water.After the trip they come back to ask people to pay attention to the environment and they accept their advice.I’m sure the story can win the hearts of many people with the interesting characters, beautiful pictures and the happy ending.Module 6
Dear Jane,Thanks for your letter.I’m very glad to talk about my hobby to you.I have a few hobbies, but my favorite one is going online.I started to go online when I was very young.Now I still enjoy it.I’ve learned a lot from the Internet and I can do a lot of things on it.For example, I can send emails and do shopping online.It helps me save time and money.I think hobbies can make us grow as a person, develop our interests and help us learn new skills.However, we shouldn’t spend all our free time on our favorite hobby.Best wishes!
Yours，John
Module 7
Dear Sir,I want to join in the summer camp.It’s a good way to improve my English and it can also help me learn more about the American culture at the same time.I like sports and pop music.I can use the Internet to get something useful.I do quite well in English and can talk with foreigners freely.As well as learning English, I want to experience life in the USA.I expect to live with an American family and take part in American life.Having meals with the family and doing some activities with them.I hope to stay in a large American family with three or four children.If they want to learn Chinese, I can help them a lot.I’m sure we’ll have a great time together.Thank you!
Yours,Meimei
Module 8
It was sunny last Saturday.I went to visit Beihai Park with my parents.There were so many people there and we stayed there for a whole day.The park is really large.It is 682,000 square meters.There are many interesting tourist spots there, such as White Pagoda, Qionghua Island and so on.The most famous one is White Pagoda.It has already become the symbol of Beihai Park.Beihai Park is one of the earliest buildings and the best gardens in the world.It is old, but beautiful.Every year, millions of visitors from different parts of the country and different parts of the world come to visit it.It has become more and more famous around the world.Module 9
My Best Friend
In our lives, we may have many friends.And good friends really mean a lot to us.One of my best friends is Mary.She is very kind and helpful.Two years ago, I came to this city with my parents.On the first day of my new school, I felt lonely because I had no friends.Just then, Mary came to talk to me.What’s more, she helped me with my English, too.Mary’s help made me feel happy and warm.As a result, we became friends soon.It was Mary who gave me the greatest help when I was new here.I think friendship is so important that it can help us feel happy and make our lives more beautiful.Module 10
My Hobby and My Future
Reading has become my hobby since I was young.I still remember I often asked my mother to tell me stories when I was a child.My mother taught me to read books at the age of five and I became interested in reading.As I grew older, my interest in reading grew.It seemed that I could communicate with the the characters in the books.With the help of reading, I became intersted in writing, too.Although I’m very busy at school, I enjoy writing about our lives at school and intersting stories around me.Many of my classmates love my stories.I hope I will write good books for the world in the futur and become a writer.I think my hobby will bring me both pleasure and success.期末2010-2011
Dear Tom,I’m very glad to hear from you.I like playing football, too, but my favourite hobby is reading storybooks.I always go over my lessons and do my homework first after I get back home every day.Then if I have enough free time, I will read my favourite storybooks.And it usually takes small pieces of time.In my opinion, good hobbies can make us grow as a person and they bring us enjoyment and success.However, we shouldn’t spend all our free time on hobbies.As a student, we must work hard at the schoolwork.I hope you will do well in your schoolwork while enjoying your favourite hobby soon.Best wishes!
Yours,Li Hua
模拟考试
According to the result, 75% students think homework is good for them.Doing homework can help them not only go over what they’ve learned but prepare for the next lessons.Also they can improve their learning skills by finishing homework.However, 25% students have different opinions.They think doing too much homework will take too much time, as a result, they can’t have a good rest and have no time for hobbies.In my opinion, it is necessary and helpful to do homework properly.I hope our
运动目标检测处于智能视频监控系统的最底层, 其检测结果的好坏将直接影响后续的处理效果, 故其技术发展和应用前景受到广泛关注。目前常用的运功目标检测算法有:背景差分法、光流法及帧间差分法。背景差分法能够较完整的提取出运动目标, 但对光照及外部条件引起的动态场景变化过于敏感;光流法是基于对光流场的估算进行检测分割的方法, 计算复杂, 须有特殊硬件设备支持, 实时性差;帧差法是根据图像中像素点的灰度差计算出运动物体的位置和形状等信息, 这种方法对于动态环境有很好的适应性, 算法简单, 易于实现, 但不能提取出较完整的运动目标。
视频图像的边缘信息抗干扰性能好, 其不易受亮度突变及噪声的影响, 因此本文融合图像边缘信息与四帧差分对运动目标进行检测, 计算简单, 实验结果表明该算法可以有效改善传统帧间差分算法出现的空洞现象。
2 融合边缘检测的四帧差分运动目标检测
2.1 预处理
众所周知, 灰度图像的处理效率高于彩色图像, 因此首先对图像进行灰度化。另外, 图像获取过程中不可避免的存在诸如斑点噪声及椒盐噪声等一系列噪声影响, 为了有效抑制噪声影响, 本文采用中值滤波法对原始图像进行滤波处理。最后本文采用直方图均衡化及梯度化操作, 使得图像轮廓更为清晰, 以增强抗干扰能力。
2.2 Canny边缘检测算法
边缘作为图像的最基本特征广泛存在于目标物与背景之间、目标物与目标物之间, 在图像处理中有着重要的作用和广泛的应用。目前常用的边缘检测算子包括Robets算子、Sobel算子、Prewwits算子、Log算子、二阶Laplace算子等。这些算子简单, 易于实现且具有很好的实时性, 但抗干扰性能差, 对噪声较敏感且边缘的精度有待提高。基于最优化算法的Canny边缘检测算子, 是先采用高斯函数对图像进行平滑处理, 再进行边缘检测, 提取的边缘线形连接较为完整, 定位准确性较高, 效果较好。本文采用此方法对序列帧图像进行边缘提取。.
Canny算子的基本原理是通过查找图像梯度强度局部的最大值的方法来得到边缘信息, 图像梯度的计算采用Gauss滤波器。为了消除噪声干扰、提高边缘检测的精度, Canny采用双阈值来提取边缘像素点。如果梯度强度不满足高阈值的检测条件, 但是与己检测出的较强边缘点相连接, 并且满足低阈值条件, 则确定为弱边缘点。双阈值法使得采用Canny算子提取边缘点更具有鲁棒性。
Canny算法的实现步骤如下:
(1) 高斯滤波:选取合适的Gauss窗函数和尺度, 用Gauss滤波器对需要处理的图像进行平滑滤波, 得到平滑图像;
(2) 梯度计算:用一阶偏导的有限差分来计算梯度的幅值和方向;
(3) 量化梯度方向:对梯度方向进行量化, 使其只有8个方向, 即0º、45º、90º、135º、180º、225º、270º、315º;
(4) 梯度非最大值抑制:对于每一个梯度强度非零的像素点, 查找沿其量化后的梯度方向的两个相邻像素点, 然后通过阈值判断为非边缘点还是候选边缘点;
(5) 双阈值判决:设定双阂值, 进一步判断候选边缘点为强边缘点、弱边缘点还是非边缘点;
(6) 边缘连接:对于弱边缘点, 如果它与强边缘点在8连通区域相连, 则该点记为弱边缘点, 同时将其边缘提取置为1, 所有标记为1的点就成了最后的边缘。
2.3 连续四帧差分算法基本原理
传统的三帧差分算法虽然能够快速检测出运动目标的轮廓, 但其所检测出的轮廓往往不连续且存在较大的空洞现象, 同时传统三帧差分算法不能完整提取目标信息。本文采用如下改进帧间差分法, 即连续四帧差分算法, 可以在一定程度上克服传统三帧差分算法的不足。设预处理后的连续四帧图像f1 (x, y) 、f2 (x, y) 、f3 (x, y) 、f4 (x, y) , 首先分别将f1 (x, y) 与f3 (x, y) 以及f2 (x, y) 与f4 (x, y) 做差分运算, 得到差分结果:
然后将差分后的结果D1与D2二值化处理后进行“与”运算, 得到结果DA=D1D2, “与”运算能够有效的克制目标重叠现象。
2.4 本文算法
本文算法融合图像边缘信息与四帧差分对运动目标进行检测, 首先采用Canny边缘检测算法快速提取连续四帧图像的边缘图像, 然后利用四帧差分算法得到较为完整的运动目标轮廓, 最后通过后期一系列处理提取出准确完整的运动目标区域。本文算法整体流程如图1所示。
2.5 形态学处理
由于干扰因素的存在, 导致我们得到的二值化图像中往往会出现噪声及空洞, 可以通过数学形态学对其进行后处理。首先采用腐蚀运算消除二值图像中的孤立噪声点, 再采用膨胀运算来填充运动目标的边缘空洞。
3 实验对比
本文采用Matlab进行仿真实验, 为验证算法的有效性, 分别利用传统三帧差分法、四帧差分法以及本文算法对同一组AVI视频序列图像进行测试。检测结果如图2所示。
分别采用传统三帧差分算法、四帧差分算法及本文算法对原始图像进行处理, 原始图像中右边白色衣服人物是静止不动的, 处理结果如图2所示。由图可知, 采用传统三帧差分算法检测到的目标轮廓不连续, 四帧差分法得到的图像目标轮廓更清晰, 内容更为丰富, 但轮廓依然不完整, 而采用本文算法检测得到的目标轮廓更加完整清晰, 结果准确。
4 结论
本文融合图像边缘信息与四帧差分对运动目标进行检测, 首先采用Canny边缘检测算法快速提取连续四帧图像的边缘图像, 然后利用四帧差分算法得到较为完整的运动目标轮廓, 最后通过后期一系列处理提取出准确完整的运动目标区域。实验结果表明本文算法能够快速准确检测出运动目标, 进一步改进了传统三帧差分目标轮廓不连续现象, 计算简单, 可满足实时性检测需求。
参考文献
[1]罗志升, 王黎, 高晓蓉, 王泽勇, 赵全轲.序列图像中运动目标检测与跟踪方法分析[J].现代电子技术, 2009.
[2]蒋钘.人体运动目标检测与跟踪关系问题研究[D].中南林业科技大学, 2013.
[3]刘辰飞.智能视频监控系统中运动目标的检测与跟踪[D].济南大学, 2012.
[4]舒欣, 李东新, 薛东伟.五帧差分和边缘检测的运动目标检测[J].计算机系统应用, 2014, 23 (1) .
[5]许录平.数字图像处理[M].北京:科学出版社, 2007.
[6]张德丰.MATLAB数字图像处理[M].北京:机械工业出版社, 2009.
[7]邵广安.视频监控中运动目标检测与跟踪方法的研究[D].南昌大学, 2013.
A)根据句意及首字母提示完成单词。
1. ——What’s the w____ like in Beijing?
——Snowy.
2. How t____! I lost my keys.
3. The beach in Beidaihe is p____ good. I like it.
4. You can get m____ from the bank.
5. Look! The children are h____ a good time in the park.
B) 根据中文提示，用单词的适当形式补全句子。
6. I’ll arrive this afternoon. Can you meet me at the ____ (飞机场)?
7. Let’s go ____ (通过) the street. The lights are green.
8. Some are reading, and ____ (其他的) aresleeping.
9. People on vacation are very ____ (放松的).
10. ——May I use your dictionary?
——____ (当然).
Ⅱ. 单项填空(共15小题，每小题1分，共15分)
1. Walk ____ the gate, and you can see a beautiful park.
A. acrossB. through
C. inD. out
2. I live in school and I often talk with my parents ____ the phone.
A. onB. in
C. throughD. as
3. You may get ____ after you see the movie. It’s interesting.
A. shyB. cold
C. relaxedD. ugly
4. My mother is a good ____. She can ____ Chinese and Japanese food.
A. cooker; cookB. cook; cook
C. cook; cooksD. cook; cooker
5. Do the students in this school ____
uniforms when they’re at school?
A. sellB. give
C. wearD. get
6. The students have ____ homework to do.
A. manyB. much
C. mostD. any
7. Dogs are ____. They are people’s ____.
A. friend; friendlyB. friends; friendly
C. friendly; friendD. friendly; friends
8. ——What ____ she ____?
——She is a nurse.
A. does; doesB. does; do
C. do; doesD. do; do
9. My grandma likes ____ stories.
A. sayingB. speaking
C. tellingD. talking
10. ——Why do you want to go to Australia?
——____ I like the beaches there.
A. IfB. So
C. AndD. Because
11. He is good at ____, but he is not good at ____.
A. runing; skating
B. run; skate
C. running; skateing
D. running; skating
12. I have two notebooks. One is red, and ____ is green.
A. otherB. the other
C. othersD. the others
13. Look! The students ____ themselves in the park.
A. enjoysB. enjoy
C. are enjoyingD. is enjoying
14. ——____?
——It’s a terrible day. It’s raining all the time.
A. Is it a fine day
B. How’s the weather today
C. What is it doing
D. How’s everything going
15. Many people are ____ the buses at the bus station.
A. waiting forB. playing with
C. coming fromD. looking for
Ⅲ. 完形填空(共10小题，每小题1分，共10分)
Dear Peter,
School is going well. I am so __1__ and I have little time to go and play __2__ the weekend. I always go to the library to __3__ newspapers or study English.
If you’re free, you can come to __4__ me. You can go straight down Center Street. __5__ you see Blue Bar, just turn left. The name of my apartment(公寓) is Victor. That’s easy to __6__. I like my house. It isn’t very big, __7__ it’s a nice home. If you come to my house, we can watch movies, or we can __8__ Chinese food. I’m often at home after 7 pm, and I __9__ at around 10:30 pm every day because I have to go to school early the next morning. You can give me a __10__ before you come.
Your best friend,
Linda
1. A. lazyB. busy
C. happyD. excited
2. A. inB. with
C. atD. from
3. A. readB. buy
C. giveD. mail
4. A. tellB. teach
C. watchD. visit
5. A. SoB. Because
C. WhenD. Before
6. A. findB. look
C. liveD. pay
7. A. andB. but
C. soD. or
8. A. sellB. clean
C. learnD. cook
9. A. get upB. go to bed
C. take a walkD. go shopping
10. A. callB. number
C. talkD. time
Ⅳ. 阅读理解(共15小题，每小题2分，共30分)
（A）
Dear Amy,
Do you like animals? I do. I love to go to the zoo. I want to tell you about my favorite animals. I like pandas. They are black and white, and they look so cute. Pandas are from China. They eat bamboo leaves. I like lions, too. They are from South Africa. They are scary, but they are beautiful. Lions eat meat. The penguins are really interesting. They are black and white and they eat fish(鱼). The elephants are interesting, too. They are very big and friendly. They don’t live in the water, but they need a swimming pool. The tigers are from India. They are very beautiful and exciting animals. They are kind of scary, too.
What are your favorite animals?Yours,Michael
根据短文内容，选择正确答案。
1. ____ are scary.
A. Lions and penguins
B. Lions and tigers
C. Elephants and tigers
D. Pandas and tigers
2. What do the penguins eat? ____.
A. FishB. Meat
C. LeavesD. Bamboo
3. Where are tigers from? ____.
A. ChinaB. South Africa
C. IndiaD. Australia
4. What does the word “water” mean? ____.
A. 水B. 笼子
C. 河流D. 展览馆
5. We can know that ____.
A. Amy likes animals
B. pandas and penguins are the same color
C. elephants need a lot of water to
drink
D. tigers and lions are very big and friendly
(B)
根据以上信息，读下面这些学生的话，他们应该加入什么俱乐部？
6. Sally: I want to visit Paris, but I can’t speak French.
7. James: I like seeing movies and I want to be an actor.
8. Tina: I will join a party and I want to show myself by singing or dancing.
9. Van: I like eating hamburgers and French fries. I want to know how to make them.
10. Tony: I can play tennis, but I can’t play it well. I want to improve it.
6. ____7. ____8. ____
9. ____10. ____
（C）
There are three people in Larry’s family. Larry’s father is a policemen. __11__ Today is Sunday. __12__ Larry and his parents come to the park. There are many people in the park. Some are taking a walk, and some are taking photos. __13__ Larry and his parents are at a table. There is some fruit on the table. __14__ __15__ His father is answering Larry’s questions. His mother is listening to them. They are very happy. This is a nice family.
根据短文内容，完成下列任务。
任务一：将下列3个句子填入短文11-13的空白处。
11. ____12. ____13. ____
A. Larry’s mother teaches English at a school.
B. Some kids are playing games.
C. It is sunny and warm.
任务二：根据汉语提示，写出14和15小题所缺的句子。
14. 拉瑞在他父母中间。
_________________
15. 拉瑞和他的父亲在讨论一本书。
_________________
Ⅴ. 句型转换(共5小题，每小题2分，共10分)
按括号内要求改写下列句子。
1. My mother works in a restaurant.(就划线部分提问)
____ ____ your mother work?
2. Mary likes singing. (改为一般疑问句)
____ ____ ____ singing?
3. Jim’s parents speak English.
(改为否定句)
Jim’s parents ____ ____ English.
4. What’s her job? (改为同义句)
What ____ ____ ____?
5. work, for, can, you, a, newspaper
(连词成句)
__________________?
Ⅵ. 翻译句子(共5小题，每小题2分，共10分)
根据汉语意思完成英语句子，每空一词。
1. 教室里有人吗？
____ ____ anybody in the classroom?
2. 如果你迟到了，我们不会等你。
____ you’re late, we won’t ____ ____ you.
3. 亨利每天搭出租车去上班。
Henry ____ ____ ____ to work every day.
4. 祝你们玩得开心！
I hope you ____ ____ ____ ____!
5. 那个小男孩正躺在床上。
The little boy ____ ____ on the bed.
Ⅶ. 情景交际(共5小题，每小题1分，共5分)
从方框中选择合适的选项补全对话(其中有两项多余)。
A: Hi, I’m Nancy! I’m your new classmate. What’s your name, please?
B: __1__ Where are you from?
A: I’m from Australia.
B: Are your parents Australians?
A: Yes. My father is a teacher. __2__.
B: What do you want to be?
A: __3__ What about you?
B: I want to be a newspaper reporter.
A: __4__
B: Because it’s an interesting job.
A: __5__
B: Well, I like it, though.
A. Why?
B. Are they busy jobs?
C. My name is Carol.
D. I want to be a policewoman.
E. What do your parents do?
F. My mother is a bank clerk.
G. But it’s kind of dangerous as a reporter.
Ⅷ. 书面表达(10分)
Emma有一个笔友Amy，她们刚认识不久，Emma想给Amy进一步介绍自己的情况。假设你是Emma，请根据下面的提示给Amy写一封50词左右的短信。
个人爱好：喜欢动物和体育
家庭成员：父亲是记者，母亲是老师
天气情况：很糟，夏天很热，冬天很冷
Dear Amy,
_____________________________________________________
_____________________________________________________
_____________________________________________________
______________________________________
新目标中考英语短语检测第26页
1.一点 2.一瓶 3.一些，几个 4.一些，一点 5.许多 6.大量的 7.大量，许多 8.一张 9.一茶杯 10.一份，一本 11.各种各样的 12.一生
13.遍及每一部分，浑身 14.遍及全世界 15.仍然 16.一直 17.一年到头 18.过一会儿 19.一次又一次 20.在。。岁 21.在…开端 22.在…未尾 23.在…排头 24.在…脚下 25.此刻 26.在同时 27.有时 28.和…一样 29.和…不一样 30.只要 31.一…就 32.事实上，其实 33.似乎
34.除…外，也，既…又 35.像平时一样
36.向…要…,请求
御花苑外国语学校2010年4月19日星期一
37.请假 38.到达 39.在吃早饭 40.在工作 41.在午夜 42.在夜里/中午 43.首先 44.最后 45.至少 46.立刻，马上 47.能够 48.生某人的气 49.对某事生气 50.醒来 51.出生 52.害怕… 53.害怕做… 54.恐怕… 55.对…感到惊讶 56.以…为依据 57.回来 58.在家 59.出去，外出 60.小心，当心，注意 61.被…覆盖 62.为…作好准备 63.对…感到惊讶 64.在进行，在上演 65.对…感到高兴
一种新的运动目标检测与跟踪算法
常用的.运动目标检测算法无法解决在摄像机运动-目标运动情况下的运动目标检测,为此提出了背景匹配法.通过相关匹配算法使背景对齐,结合帧间差分技术有效地将运动目标提取出来;然后利用卡尔曼预测器对运动目标在图像中的位置进行预测,结合增量式带死区的PID控制算法,控制摄像机对准目标.实验结果表明,匹配块的选择加快了处理速度和提高了算法的稳定性,卡尔曼预测器使得跟踪更为平稳可靠.本方法具有简单、通用、抗噪等特点.
作 者：潘锋 王宣银 向桂山 梁冬泰 作者单位：浙江大学,流体传动及控制国家重点实验室,浙江,杭州,310027刊 名：光电工程 ISTIC PKU英文刊名：OPTO-ELECTRONIC ENGINEERING年，卷(期)：32(1)分类号：V556 TP391.4关键词：目标检测 目标跟踪 背景匹配 卡尔曼预测器 PID控制
八年级数学参考答案之四
一、填空题（每小题5分，共40分）
1、x≠
12、(a+b)(a-b)3、7、618、9014、y=-x 等
5、-
26、减少m3
二、选择题（每小题5分，共40分）
9、D10、C11、A12、D13、A14、D15、B16、B
三、解答题（每小题10分，共40分）
117、解：原式=x18、解：设原计划每天铺设x米，则实际每天铺设（1+25%）x米，依题意得3000300030，解得x=20，经检验 x=20是原方程的根，x(125%)x
答：略。
14xy=2，∴y =。2x
m120、解：由题意得 3=x0+m，3=，解得x0=1，m=2，x
3∴一次函数的解析式为y=x+2，反比例函数的解析式为y=。x21、解：ΔABC 是直角三角形，理由如下：
7∵a+b=4，c=3，∴(a+b)2 =42，∴a2+2ab+b2=16，又ab=，∴2ab=7，∴a2+b2=9，2
又c2=9，∴a2+b2=c2，∴ΔABC是直角三角形。
随着科学技术快速发展, 视频监控系统得到越来越广泛的应用, 其正向数字化、网络化、智能化方向发展。
运动目标的检测与跟踪是智能视频监控系统中最基础的两项核心技术, 它们是后续各种高级处理, 如行为分析、事件检测、行为识别、智能预警、视频图像压缩编码和语义索引等高层次视频处理和应用理解的基础, 也是视频监控技术智能化和实时应用的关键。传统的运动目标检测与跟踪技术往往只适用于特定的场景, 对场景中各种复杂变化难以适应, 且在实时性和准确性方面很难达到平衡。针对上述问题, 本文在固定摄像机视频监控系统前提下, 将运动目标检测分为背景相减法和基于目标特征检测法进行简单讨论, 然后使用一种基于区域划分的运动目标检测方法对简单场景中运动目标进行检测, 并进行实验分析, 为提高智能视频监控系统对运动目标检测的准确性和实时性做出有益探索。
1 视频图像中运动目标的一般检测方法
智能视频监控具有传统模拟视频监控无法比拟的优点, 符合视频监控发展趋势。近年来在各个领域都有极其广泛的应用, 如停车场、交通路口、高速公路以及重要路段的交通流量监测与违章事件监控;应用到军事和国家重点安全部门, 如军事基地、监狱、博物馆、边境等防范意外事件或恐怖分子的袭击;还有敏感公共场合, 如天安门广场、火车站、银行及社区和家庭的视频监控等。人们从这些监控系统中能得到大量的监控数据, 并希望在安全防范和决策支持方面, 系统能够提供智能化帮助。运动目标的检测是视频监控系统中智能化分析的基础, 本节将对运动目标检测的背景相减法和基于目标特征检测法进行简单讨论, 提出它们应用中存在的问题。
1.1 背景相减法
背景相减法 (Background Subtraction) 是目前运动检测中最常用的一种方法, 它是利用当前图像与背景图像的差分来检测出运动目标的一种技术, 包括预处理、背景建模、前景检测和后处理四个步骤。
实际上, 背景的建模是背景相减法的技术关键。目前, 人们基本都在研究如何建立更加实用和健壮的背景模型, 使背景能随时间重建和更新, 减少光照变化、背景扰动等对运动目标检测的影响。
1.1.1 时间差分法
时间差分法[1] (Temporal Difference) 是利用视频序列中当前两帧或当前帧与前某帧的差分来提取出图像中的运动区域。直接利用t-1时刻的视频帧作为t时刻背景帧的方法又叫帧差法 (Frame Difference) 。文献[2]中还有一种方法是先以帧差法求出当前t时刻与t-1时刻的变化区域, 然后将t时刻背景的非变化区域以当前帧的像素代替, 这时背景图像大部分与当前帧一样, 再以这个背景图像去和t+1时刻的视频帧进行差分来得出运动目标区域, 这种方法实际上是在帧差法中引入可变背景, 而不是简单地利用前一视频帧作为背景。
可见, 时间差分法具有较强的场景变化适应能力, 背景不随时间积累, 且更新速度快。但由于它得到的是两帧同位置的像素“或”, 因此, 判定颜色一致的、比较大的运动目标的内部像素时比较容易困难, 以至在运动实体内部产生空洞现象。因此, 要得到好的性能, 背景帧选择必须考虑运动目标的速度和帧率。对于快速运动目标, 背景帧应该与当前帧靠近, 甚至就是前一帧, 否则很有可能把同一目标检测为两个目标。
1.1.2 中值滤波法
中值滤波和均值滤波 (Median/Average Filtering) 思想上差不多, 这里将它们归为一类, 它们是用得最多的背景图像构建方法[3]。其算法思想是先建立一个视频流滑窗用来缓存L张视频帧, 然后把缓存中所有视频帧同位置像素的平均值或中值作为背景中该处像素的值。
可见, 中值滤波和均值滤波需要L倍帧大小的内存, 它们适用于场景目标不大且运动连续的情况, 对于大而慢的运动目标检测, 则需要相应增大滑窗。为了减少内存, 另一种改进的方法称为运行期均值法RA[3] (Running Average) , 其思想是通过引入学习率λ来体现背景图像对场景变化的响应, λ通常取0.03, λ越低, 前景的变化越不会影响背景。
1.1.3 混合高斯模型
对于室外场景, 因为经常有摇动的树干/叶等, 监控背景通常不会完全静止。那么, 在没有运动目标的情况下, 对于相同位置的像素, 在视频序列的某些帧中表示地面, 而在另一些帧中则表示树叶。这类背景模型可以采用混合高斯模型GMM[4,5] (Gaussian Mixture Model) 建模, 在文献[4,5]中给出了GMM的原理。
GMM适用于室外经常存在树干/叶等背景扰动的场景, 但其收敛速度缓慢、时间复杂性偏高。而针对收敛速度的改进模型[6], 只是提高其初始化的收敛速度, 初始化模型与混合高斯模型相互独立, 在初始化后的目标检测过程中, 当光照突变等导致背景发生改变时, 这些改进模型对混合高斯模型收敛速度的改进效果较差;而针对时间复杂性的改进模型[7], 则主要是利用每个像素对应高斯分布个数的自适应调整来降低时间复杂性。
1.2 基于目标特征的运动目标检测
除背景相减法外, 还可利用运动目标的特征信息来检测目标, 如运动目标颜色、纹理、边缘、形状和运动信息。但该方法常常涉及到匹配、光流分析等复杂的运算, 难以满足实时处理的条件, 且一般对先验知识的依赖性较强, 导致其鲁棒性受到一定影响。
比较以上两种方法, 背景相减法仅识别那些有变化的像素, 可以提供较好的检测效果, 且时间复杂性小。因此, 为了实现实时自动的检测运动目标, 应当选择背景相减法或者以背景相减法为主, 并融合目标特征等先验知识的方法来实现运动目标的检测与提取。
2 针对简单场景运动目标检测的MODBAP方法
通常人们将固定摄像机视频监控系统的应用场景分为简单场景和复杂场景两类, 当摄像机安装的位置固定时, 其监控区域不再改变, 即应用场景已经确定。本文主要讨论研究的是简单场景的固定摄像机视频监控, 目前比较流行的方法是采用运行期均值法RA检测目标, 但RA对背景相减法阈值的设定和场景光照的变化比较敏感。文献[8]采用光流法构建精细的背景模型, 并采用彩色信息进行阈值自适应分割, 该方法虽然适用于场景光照突变的情况, 但时间复杂性显著提高, 其实时性得不到满足。那么针对简单场景, 这里使用一种基于区域划分的运动目标检测方法MODBAP (Moving Object Detection Based on Area Partition) , 该方法对阈值设定和光照条件变化不敏感, 时间复杂性低, 并能更有效的抑制噪声和检测出完整的运动目标。
MODBAP基本流程如图1所示。
监控视频处理过程中, 视频背景的地形和道路等信息可以被充分利用, 因此, 可以人工确定视频背景中的目标出现区域, 即运动目标进入视频背景的区域;相对于整个视频背景区域, 目标出现区域要小得多, 因此, 只在目标出现区域内利用Canny算子[9]提取的边缘信息检测运动目标, 这对时间效率上影响较小;根据视频序列快速的帧率, 运动目标在相邻两帧的差异很小, 因此, 当运动目标离开目标出现区域, 可以在背景相减法获得的差异图像上, 利用目标运动的连续性, 根据运动目标在前一帧被检测出的区域, 自适应检测运动目标在当前帧的区域。根据上述思路, 下面给出MODBAP的算法步骤。
2.1 目标出现区域划分及背景图像获取
在固定摄像机安放位置确定时, 监控区域不再改变, 那么根据视频背景划分的目标出现区域在一定时间段内也不再改变。由于受到地形、建筑布局等限制, 运动目标一般只能从特定的区域进入视频背景。通常运动目标会从监控区域边界进入视频背景, 如图2边界处的闭框即为划定的目标出现区域, 当监控区域有遮挡物时, 则在遮挡物的边界处也可划定目标出现区域。下面以图2 (a) 对应监控视频图像为实例, 设计目标出现区域内 (外) 运动目标的检测步骤。
这里采用简单的运行期均值法RA获取背景图像, 即更新方式为:
式中:It (x, y) 和Bt (x, y) 分别表示t时刻当前图像和背景图像中位置 (x, y) 处的像素亮度;Bt+1 (x, y) 表示t +1时刻背景图像中位置 (x, y) 处的像素亮度;背景学习率λ通常取值为0.03。
2.2 目标出现区域内运动目标的检测
如图3 (a) 为图2 (a) 对应监控视频中的一帧图像, 有目标进入目标出现区域内, 图3 (b) 为图3 (a) 对应的背景图像。因为简单场景的背景纹理平滑, 所以背景图像中的边缘信息较少。因此, 利用当前图像与背景图像边缘信息的对比, 可以实现目标出现区域内运动目标检测。目标出现区域内运动目标的检测步骤描述如下 (将“0～255”二值图像I中像素x的值用I (x) 表示) :
步骤1:利用Canny算子提取当前图像和背景图像在目标出现区域的边缘, 分别用二值图像I1和I2表示, 如图3 (d) , 图3 (e) 所示, 将I2用3×3结构元素的膨胀算子处理一次后得到I3, 此时, 对∀像素x, 若I1 (x) =255且I3 (x) =0, 则I4 (x) =255, 否则I4 (x) =0, 可获得如图3 (f) 的二值图像I4, 即去除背景边缘后的运动目标边缘。设定阈值T, 如果I4中值为255的像素个数少于阈值T, 则认为目标出现区域内没有目标, 否则进行下一步处理。
步骤2:如果I4中运动目标边缘是完整闭环, 则可以通过填充获得完整的运动目标, 但由于噪声和I3 (背景边缘) 的影响, I4中边缘没有形成闭环, 而是由宽度为1个像素的连通曲线组成, 图4 (a) 是图3 (f) 中的一个连通曲线, 图4 (a) 中共有3个红色连通域, 每个连通域对应连通曲线的一个端点, 图4 (b) ～ (d) 中的红色正方形是以图4 (a) 的连通曲线中某一像素为中心的3×3结构元素。图5 (a) ～ (c) 分别对应给出了图4 (b) ～ (d) 中红色正方形的局部放大, 由图5 (a) ～ (c) 可知:当且仅当连通曲线中与红色正方形相邻的像素 (除绿色像素) 在同一连通域时, 绿色像素为端点像素。由此规律, 可以找出I4中所有连通曲线的端点 (如图3 (g) 的红色像素) 。为获得封闭的运动目标边缘, 需要消除如图3 (g) 中的端点。图5 (d) 截取了图3 (g) 中的一个连通曲线L, 以L中的一个端点为中心不断向外寻找相邻的像素, 如图5 (d) 中相互嵌套的浅灰色和深灰色闭环, 直到与L以外其他连通曲线相交, 相交部分如图5 (d) 中的绿色像素, 然后将红色端点与绿色像素相连, 如图5 (d) 中的黄色线段, 这样即可消除如图3 (g) 中的端点。考虑到噪声等的影响, 设定阈值D, 当如图5 (d) 中浅灰色和深灰色闭环数目超过D还没有与L以外其他连通曲线相交时, 则对应的端点不与其他任何连通曲线相连。将图3 (g) 中端点进行连接后, 得到如图3 (h) 的结果。
步骤3:由于内部边缘将运动目标分割成多块区域, 因此采用对运动目标外部进行填充的方式, 如图3 (i) 中的灰色区域即填充区域, 然后将未被填充的区域作为运动目标区域如图3 (j) , 为了消除噪声的影响, 对图3 (j) 进行3×3结构元素的腐蚀和膨胀算子处理各一次, 得到完整运动目标, 如图3 (k) 所示。图3 (c) 即图3 (a) 在目标出现区域内检测出的目标。
2.3 目标出现区域外运动目标的检测
在目标出现区域划定带方向的红色越界线, 如图6 (a) 所示, 然后利用目标出现区域内的检测方法得到如图6 (b) 的结果, 图6 (b) 中目标的中心点越过越界线, 因此, 图6 (a) 的下一帧图6 (d) 采用目标出现区域外的检测方法。图6 (d) 与图6 (a) 间隔0.1 s, 相互差异不大。目标出现区域外运动目标的检测步骤为:
步骤1:当前图像减去背景图像的绝对值构造如图6 (e) 的差异图像F, 用F (x) 表示差异图像F中像素x的亮度值;设像素x的4邻接N4 (x) = {x1, x2, x3, x4}, 如图6 (f) 所示, 对差异图像F中的像素进行分类标记:自顶向下、自左到右扫描差异图像F一次, 设当前扫描至像素x, 计算
步骤2:设前一帧检测出的运动目标区域为M, 非运动目标区域为
2.4 算法实验结果及其讨论
实验使用监控视频如图7 (a) ～ (c) 所示, 在Windows操作系统下, 算法采用VC++6.0编写。
首先采用运行期均值法RA检测运动目标, 其中RA背景相减法阈值的自适应性采用文献[10]的自适应阈值Th获取方法:
(1) 选择平均亮度值作为Th的初始估计值;
(2) 用Th分割生成两组像素:G1由亮度值大于Th的像素组成, G2由亮度值小于等于Th的像素组成;
(3) 计算G1和G2中像素的平均亮度值μ1和μ2;
(4) 计算新阈值Th= (μ1+μ2) /2;
(5) 重复步骤 (2) ～ (4) , 直到逐次迭代所得的Th值之差不变为止。
然后将MODBAP与RA进行对比, 图7 (d) ～ (f) 为MODBAP的实验结果, 图7 (g) ～ (i) 为自适应阈值RA的实验结果。通过实验对比发现:MODBAP能更加完整的检测出运动目标, 且对光照的敏感性低于自适应阈值RA;MODBAP对输入阈值参数不敏感, 阈值取值在合理范围内时, 对实验结果没有明显影响;MODBAP不需要构造精细的背景模型, 采用clock () 函数的100帧均值测试CPU执行时间, 可以发现MODBAP对每帧图像的平均处理时间为0.023 s, 可以有效解决简单场景的实时运动目标检测问题。
3 结 语
在固定摄像机视频监控系统简单场景的运动目标检测中, 本文使用一种MODBAP方法, 改善了传统运动目标检测技术的不足, 提高了目标检测准确性和光照低敏感性, 并且满足视频序列实时性, 保证了监控系统后续处理能有充裕的时间。该方法准确、稳定、快速, 在诸多固定摄像机视频监控系统具有很高的实用价值。
摘要：随着视频监控系统的普及化和庞大化, 传统的运动目标检测技术难以适应检测场景的各种变化, 局限性越来越突出, 且实时性和准确性难以平衡。为有效解决传统的运动目标检测技术中出现的问题, 以固定摄像机视频监控系统为研究背景, 使用一种基于的运动目标检测方法, 提高了运动目标检测准确性, 降低了对光照的敏感性, 并保证了实时性。
关键词：智能视频监控,运动目标检测,背景相减法,区域划分
参考文献
[1]LIPTON A, FUJIYOSHI H, PATIL R.Moving targetclassification and tracking from real-ti me Video[C]//Pro-ceedings of IEEE Workshop on Applications of ComputerVision.USA:Princeton, NJ, 1998:8-14.
[2]艾海舟, 吕凤军.面向视觉监视的变化检测与分割[J].计算机工程与应用, 2000, 37 (5) :75-77.
[3]CUCCHIARA R, PICCARDI M, PRATI A.Detectingmoving objects, ghosts and shadows in video streams[J].IEEE Transactions on Pattern Analysis and MachineIntelligence, 2003, 25 (10) :1337-1342.
[4]STAUFFER C, GRI MSON W E L.Learning patterns ofactivity using real-ti me tracking[J].IEEE Transactions onPattern Analysis and Machine Intelligence, 2000, 22 (8) :747-757.
[5]STAUFFER C, GRI MSON W E L.Adaptive backgroundmixture models for real-ti me tracking[C]//Proceedings ofIEEE Conference on Computer Vision and Pattern Recogni-tion.Colorado, USA:Fort Collins, 1999, 2:246-252.
[6]SUN Yun-da, LI Ming, WU Wei, et al.Background modelinitialization in moving object detection with shadow eli mi-nation[C]//Proceedings of the 7th International Conferenceon Signal Processing.China:Beijing Jiaotong University, china, 2004, 2:1288-1291.
[7]ZORAN Zivkovic.I mproved adaptive gaussian mixturemodel for background subtraction[C]//Proceedings of theInternational Conference on Pattern Recognition.Nether-lands:Amsterdam University, 2004, 2:23-26.
[8]魏志强, 纪筱鹏, 冯业伟.基于自适应背景图像更新的运动目标检测方法[J].电子学报, 2005, 33 (12) :2261-2264.
[9]CANNYJ.Acomputational approach to edge detection[J].IEEE Transactions on Pattern Analysis and Machine Intelli-gence, 1986, 8 (6) :679-698.
关键词：FPGA；目标检测；视频处理；背景差分法
中图分类号：TP391.41文献标识码：A文章编号：1007-9599 (2011) 06-0000-01
Moving Target Real-time Detection of FPGA-based Static Background
Chen Quanjin,Zhang Yide
(School of Optoelectronic Information,University of Electronic Science and Technology of China,Chengdu610054,China)
Abstract:A video detection system of moving objects based on FPGA and background difference algorithm is designed.and introduces the realization in detail.The system processing speed has been enhanced greatly,and the moving object can be detected in realtime and accurately under the staticbackground.
Keywords:FPGA; Object detection;Video processing;Background differencing
一、系统总体方案设计
系统主要分为视频采集、视频格式转换、视频预处理、视频运动目标检测、VGA视频显示和系统报警六个部分。图1为系统的整体总体框图。
（一）各部分模块的功能
1.视频采集：将模拟视频数据通过解码芯片转换为数字视频数据。2.视频格式转换：将解码后的视频数据转换成RGB格式。3.视频的预处理：包括将RGB数据存储在片外SDRAM中，图像重构和存储背景。4.视频移动目标检测：从存有当前视频图像和背景图像的SDRAM中读出背景图像和当前图像，根据背景差分算法，将背景差进行图像去噪和形态学处理，找出移动目标。5.VGA视频显示：将检测后的结果实时的在显示器上显示出来。6.系统报警：当检测到运动目标时，系统发出报警信号。
（二）系统硬件电路设计
系统采用CMOS模拟摄像头WV-GP240，解码芯片ADV7180进行视频采集，将模拟视频信号转换成FPGA能够处理的数字视频信号，FPGA作为核心处理器，负责视频的格式转换，预处理以及移动目标的检测功能，当有移动目标出现时，发出警报。差分后图像或背景图像或当前图像经数模转换芯片ADV7123转化为模拟信号，通过VGA接口显示在显示器上。图2为系统硬件结构图。
二、系统方案的设计实现
（一）视频处理模块
为了对完整的视频图像进行处理，必须对视频图像进行缓存：在SDRAM中开辟两个大小为一幅视频图像的存储区，采用ping-pong操作，当在传输和存储重构当前帧图像的时候，对上一帧图像进行后续处理。如此反复交错的进行存储重构和处理，从而提高系统的吞吐量和处理速度。
（二）运动目标检测模块
本系统采用背景差分法。算法实现的流程如图3。将当前帧图像和背景图像的灰度进行背景差计算。背景差分算法实现过程为：将当前帧图片和背景图片同时送入背景差分模块中，对两幅图像中对应的像素的相关分量进行相减运算并取绝对值。当运算结果小于某个阈值T时，认为当前帧图像和背景图像在该像素点上相同，将像素的灰度值置为255；当运算结果大于阈值T时，认为当前帧图像和背景图像在该像素点上发生了变化，表示有移动目标出现，将像素的灰度值置为0。为了使检测到的移动目标区域充实准确，引入了数学形态学处理中的腐蚀、膨胀算法。运用腐蚀算子去除孤立的噪声点，运用膨胀算子填补目标区域的空洞，这样既消除了图像中分散的碎杂噪声点，又保留了目标的边缘特征。
三、验证和测试
本设计采用Altera公司Cyclone II系列FPGA芯片（EP2C70）作为系统的硬件平台。以上算法均采用Verilog硬件描述语言设计。图4为实验结果，a为背景图，b为当前帧图像，c为背景差值图像，d为差值图像二值化处理后的图像。
由实验结果可以看出：在简单静态背景下，利用背景差分算法能够有效的检测出完整的运动目标，检测效果比较好。能够充分检测出运动目标的形状和姿态等。采用的是FPGA作为处理器，运算速度非常快，采集视频图像时采用ping-pong结构，采集过程和目标检测采用流水线处理结构，实现了目标检测的实时性。
参考文献：
[1]孙景琪.视频技术与应用[M].北京:北京工业大学出版社,2003
[2]贾国锋,韩磊.基于FPGA的视频颜色空间转换电路设计[J].系统仿真学报,2006,8:110-112
【《目标检测》作文】推荐阅读：
银川高二检测语文作文11-18
金属检测 金属成分检测05-23
目标愿望作文06-14
年度试验检测工作总结-检测检验10-16
入侵检测技术发展方向入侵检测12-16
坚定自己的目标作文05-23
朝着目标前进的作文06-17
小目标作文400字07-19
【精华】理想与目标作文07-25
实用的实现目标作文10-10
