第一篇:《目標檢測》作文
一、提建議:《目標檢測》M1 1).Dear Frank,There are a lot of things you should do.You should talk about it with your classmates.Maybe they just played a joke on you.Or maybe you should write a letter to your best friend if you think it is not easy for you to talk about it with others.Anyway, always be friendly to others and always smile to others!It is the best way to be popular!
Mary 2)(補充另一篇)...It's a good idea to watch English films and listen to real English songs.Try to read English newspapers and listen to English radio.They are good for your English.Try not to be shy when you start a conversation in English.Do you find it difficult to remember the English words? You should remember eight or ten words a day.How about placing the words on the wall or saying the English names for everything you see? Work hard, and you can learn English well.二、談論愛好:《目標檢測》M5 1)
Dear Miss Liu,My name is Lily.I’m 13years old.I’m from Class 1, Grade 8.I can play the violin.I began to play the violin at the age of 5.Now I can play it very well.I often take part in some competitions and I have won some prizes.I like music because I think music is very beautiful.It makes me happy.I also learn a lot from music.It improves my memory a lot.So I can remember English words easily.I hope I can play in our school orchestra.I believe I can do well.Yours,Lily
2)(補充另一篇)I like music very much because it can give me power when I am sad.People from different countries all like music very much.Music is very important for us.If there is no music is very important for us.If there is no music in the world, our life won't be so interesting.Pop music is my favourite.I think it sounds nice and it can make me feel relaxed.Jay Chou is my favourite singer.He is one of the most popular singers in Asia.三、介紹人物:《目標檢測》M7
My friend Linda was born in Canada.She is 13 years old.She is a tall and thin girl with long hair.She is lovely and friendly.She has many hobbies, such as dancing and pop music.She also likes shopping and swimming, and reading is her favorite one.She usually invites me to go to the library at the weekend with her.There we read books and have fun together.This is my best friend.I love her.四、談論規定:《目標檢測》M12 Miss White, it is a notice to the audience.It says that we should walk into the cinema hall fifteen minutes before the movie starts.We mustn’t smoke or talk loudly in the cinema hall.We mustn’t take food or drinks into it.We can’t use the mobile phone after the movie is on.We can’t use the camera.(補充另一篇)Dear Jack, I am very happy that you will come to visit China.I'll tell you some customs in China.You must walk on the right.When you meet other people, you should say hello to them instead of bowing.When you eat, you must use chopsticks.When you have dinner with a Chinese family, the host often asks you to eat more food.You don't need to eat it up.After the dinner, the host often makes tea for you.It's delicious.Before you leave the house, you should say goodbye to the host politely.Waiting for you!Yours, Zhang Tao
第二篇:點目標檢測
點目標檢測問題是成像制導領域中的一個關鍵技術 ,國內 外對此提出過許多算法。文獻[1 ,2 ]對傳統算法進行介紹與綜 述。總體而言 ,實時魯棒的小目標自動檢測技術尚未完全突 破 ,仍然是當今世界高技術領域的熱門研究課題。目前這些傳 統算法為了在低信噪比條件下實現對點目標較高的檢測概率 , 往往需要較大的計算量和存儲量。這些限制了具體算法在空 空導彈成像導引頭上的應用 ,從而影響了成像導引頭獲得的更 大的截獲和跟蹤距離。
近年來 ,國內外學者利用數學形態學理論對點目標檢測問
題進行了研究 ,并取得了一些成果。J.Barnett 在文獻[ 3 ]中對 形態學方法與線性空間濾波方法進行了比較;Li Jicheng(李吉 成)等在文獻[4 ]中提出利用目標圖像尺寸的先念知識選擇結 構元素消除背景而保留目標的方法;Horak文獻[5 ]中提出了基 于灰度形態學 T op2hat 變換的弱目標檢測方法 ,并在不含目標 的背景圖像上構造所需要的結構元素;Rivest、羅賢龍、Zhu Zhen fu等人在文獻[6~8 ]也分別對基于形態學的方法進行了 相應研究;熊輝等人在文獻[9 ]中提出新的思路 ,其算法基于形 態學膨脹累加 ,實現對高空背景情況下抖動目標的檢測 ,并對 抖動為1~2個像素的點目標進行5幀累加仿真。前人的研究 表明 ,形態學方法對點目標的檢測具有很大的潛力。為此 ,我 們在對成像導引頭進行分析的基礎上 ,對形態學方法應用于成 像導引頭進行研究。
由于空空導彈所攻擊的目標具有很高機動能力和很寬的速度范圍 ,所以其性能指標相對于其他紅外系統有較大區別。取典型凝視成像導引頭參數 ,進行仿真研究:幀頻為 100f/ s ,瞬 時視場為3× 3° ,搜索角速度為 15° / s ,探測器陣列為 128 ×128。經過計算 ,可知在一幀時間內 ,導引頭由于搜索而掃描過的角 度為0.15° ,即6.4 個像素。在 10km距離處 ,由導彈目標相對 運動引起的 ,目標圖像在幀間位移1.55像素。這樣 ,在對目標 截獲時 ,由于導引頭搜索和目標運動造成的目標圖像總的幀間 位移最大值約為8 像素。仿真表明 ,在導彈發射后 ,導彈在目 標跟蹤階段 ,目標圖像幀間位移約為 1~10 像素(導彈失控
前)。因此 ,目標圖像幀間位移在整個導彈點目標搜索、截獲與 跟蹤過程中變化范圍可認為在1~10像素。
仿真表明 ,利用現有基于形態學背景預測的方法進行點目
標檢測和識別時 ,對背景均勻或起伏緩慢的情況效果很好。但 對于亮度梯度高的有云天空背景和地面背景 ,算法常常造成誤 判。分析其原因 ,發現往往跟蹤的是濾波后殘留的高亮度背景 邊角。
第三篇:八年級下冊英語目標檢測作文(xiexiebang推薦)
Module 1
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
模擬考試
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
teachers should give us some creative homework , such as making posters with pictures, dreaming up new stories and giving interviews to the people around us.
第四篇:運動目標檢測方法總結報告
摘要
由于計算機技術的迅猛發展,使得基于內容的視頻信息的存取、操作和檢索不僅成為一種可能,更成為一種需要。同時,基于內容的視頻編碼標準MPEG-4和基于內容的視頻描述標準MPEG-7正在發展和完善。因此提取和視頻中具有語義的運動目標是一個急需解決的問題。運動目標提取和檢測作為視頻和圖像處理領域的重要研究領域,有很強的研究和應用價值。運動檢測就是將運動目標從含有背景的圖像中分離出來,如果僅僅依靠一種檢測算法,難以從復雜的自然圖像序列中完整地檢測出運動的目標。較高的檢測精度和效率十分重要,因此融合多種檢測方法的研究越來越受到重視。本文介紹了幾種國內外文獻中的經典的視頻運動目標的檢測和提取算法,并對各種方法進行了評價和總結。首先介紹了基本的運動目標檢測的基本知識和理論,然后介紹了基本的幾種目標檢測方法及其各種改進方法。對今后的運動目標檢測提取的相關研究提供一定的參考。
關鍵詞:運動目標檢測 光流法 幀差法 背景建模方法
摘要 i
ABSTRACT Because of the rapid development of computer technology, it is possible to access, operate and retrieve the video information based on the content of the video.At the same time, based on the content of the video coding standard MPEG-4 and content-based video description standard MPEG-7 is developing and improving.Therefore, it is an urgent problem to be solved in the extraction and video.Moving object extraction and detection is a very important field of video and image processing, and has a strong research and application value.Motion detection is to separate moving objects from the image containing background, if only rely on a detection algorithm, it is difficult to from a complex natural image sequences to detect moving target.Higher detection accuracy and efficiency are very important, so the study of the fusion of multiple detection methods is becoming more and more important.In this paper, the detection and extraction algorithms of the classical video moving objects in the domestic and foreign literatures are introduced, and the methods are evaluated and summarized.Firstly, the basic knowledge and theory of basic moving target detection is introduced, and then the basic method of target detection is introduced.To provide a reference for the research on the extraction of moving target detection in the future.Keywords: Visual tracking
Optical flow method
Frame Difference
Background modeling method
ii ABSTRACT
目錄
摘要...................................................................................................................................0 ABSTRACT.....................................................................................................................1 第一章 緒論....................................................................................................................3
1.1 研究背景及意義................................................................................................................4 1.2 研究現狀............................................................................................................................4
第二章 經典的運動目標檢測算法................................................................................5
2.1 光流法................................................................................................................................5 2.2 幀差法................................................................................................................................5 2.3 背景差分法........................................................................................................................7
第三章 改進的運動目標檢測算法................................................................................9
3.1 改進的三幀差分法............................................................................................................9 3.2 幀間差分法與光流法結合..............................................................................................10 3.3 改進的背景建模算法......................................................................................................11
第四章 總結..................................................................................................................13 參考文獻:....................................................................................................................16
目錄 i 2 運動目標檢測方法研究總結
第一章 緒論
1.1 研究背景及意義
近幾十年來,在科學技術飛速發展的條件下,視頻與圖像處理技術不斷提高,各種各樣的視頻監控產品已經走入了人們的視野,并且在給我們的生活帶了很多方便。視頻監控系統的研究技術涉及到視頻圖像處理、計算機視覺、模式識別以及人工智能等科學領域。視頻監控系統多數要求監控人員長期盯著監控屏幕,進行人為的分析判斷,這樣容易因為監控人員的疏忽造成重要信息的遺漏。為此,人們開始將計算機領域的相關技術引入到視頻監控系統中,形成智能監控系統[1,2,3,4]。智能監控系統可以在沒有人為干預的條件下,利用計算機視覺的相關技術來對視頻序列圖像進行智能的分析,實現對運動目標的檢測、跟蹤、分類和識別等。現在,運動目標檢測技術不僅應用在發電站、商場、銀行、民宅、廣場和火車站等公共場所的智能監控系統中,而且在其他的領域也有十分廣泛的應用。
1.2 研究現狀
目前,基于視頻的運動目標檢測算法主要有三種:光流法、幀差法和背景差分法,這三種算法都有各自的優缺點。1981 年,Horn 和 Schunck 通過將二維速度場與圖像灰度相聯系,從而引入了光流約束方程,得到一個計算光流的基本方法[5]。Meyer 等人[6]在對光流法進行了深入研究的基礎上,提出在光流場中采用基于輪廓的跟蹤方法,該方法在攝像機運動的情況下能夠有效的對運動目標進行檢測和跟蹤。Barron 等人[7]通過使用簡單而有效的門限,先分割圖像,再計算光流,通過消除雜亂的背景光流來得到較好的目標光流。Roland 等人[8]利用相鄰幀差,通過局部閾值的迭代松弛技術實現圖像邊緣的光滑濾波。甘明剛等人[9]提出一種三幀差分和邊緣信息相結合的運動目標檢測算法,該算法有效地改善了一些情況下幀間差分法會出現“雙影現象”的問題。郝豪剛和陳佳琪等人[10]提出五幀差分法和景差分法相結合的運動目標檢測算法,該算法利用背景差分法和幀間差分法性能上的互補來得較好的檢測結果。背景差分法有均值法、中值法、核密度估計法、Surendra 背景更新、單高斯模型和混合高斯模型等,從 20 世紀以來,相繼出現 第一章 緒論 1 了一批批成熟的背景差分法,Wren 等人[11]提出了單高斯模型,該方法在單一背景下能夠獲得較好的檢測結果,但是不適合復雜背景。Stauffer 等人[12]在單高斯模型的基礎上提出了混合高斯模型,混合高斯模型在外界環境比較復雜的條件下仍然可以得到很好的檢測效果。左軍毅等人[13]提出時間平均模型和混合高斯模型雙模式切換式的運動目標檢測算法。除了以上三種的算法外,還有一些學者嘗試采用其他的算法進行運動目標檢測,例如,郝志成和吳川等人[14]提出的基于穩定矩陣的動態圖像運動目標檢測算法,該算法通過在短時間內自動的感知背景變來快速的建立背景模型。近年來,越來越多的研究機構和學者都參與到基于視頻的運動目標檢測的研究之中,并提出很多有效的、新穎的方法。但是仍存在一些問題善待提高,所以找到一種檢測精度高、魯棒性好的運動目標檢測算法依然是我們為之努力的方向。
運動目標檢測方法研究總結
第二章 經典的運動目標檢測算法
2.1 光流法
空間中物體的運動可以用運動場來描述,同樣可以通過序列圖像中不同圖像的灰度分布差異體現圖像平面變化,對比空間中的運動場,體現在圖像上表現為光流場。在運動的某一個時刻,為圖像中的各個像素點賦一個速度的矢量,這樣就成為了一個圖像的運動場。由于空間物體上的點與圖像上的點通過投影關系可以一一對應,則根據各個像素點的速度矢量的變化特征可以對圖像進行動態分析[16]。當圖像中沒有目標運動時,在整個圖像區域中光流矢量的變化是連續的;而當圖像中有運動目標時,圖像的背景和目標就會有相對的運動,那么目標運動所形成的速度矢量必然和鄰域背景速度矢量不同,由此能夠檢測出運動目標的位置。光流法利用圖像的灰度信息的變化從序列圖像中計算出速度場,然后加上一些約束條件,從而推出運動目標的運動參數和物體結構[17]。
光流法事先不需要知道場景的任何信息,就可以準確的計算出運動物體的速度。它不僅能應用于靜態背景下的運動目標檢測,而且可以用于攝像機運動的情況,實現動態背景下的運動目標檢測。它的缺點是:光流法的特點是要進行迭代運算,精度越高需要的計算量就越大,因此,光流法的計算量大,運算時間長,是一種比較耗時的算法,很難滿足工程上對實時性的要求;光流法的抗噪性能差,例如,當光照發生變化時,即使沒有運動發生,光流仍然存在,會被誤檢測為有目標運動,同時,如果缺少足夠的灰度級變化,目標運動物體很難被檢測到;當三維物體的運動投影到二維的圖像時,亮度會有變化,從而導致通過光流約束是計算不出平面某點的圖像速度流;使用光流法對運動目標進行檢測,需要特定的硬件設備的支持。已經有一些學者針對光流法所存在的缺點進行了改進,相信未來光流法能夠得到更好的實際應用。
2.2 幀差法
幀差法[18]又叫時間差分法,它通過將視頻序列圖像中的當前幀與相鄰幀所對 第二章 經典的運動目標檢測算法 3 應的像素點的灰度值進行比較,然后找到差異,進而檢測出運動目標[19]。在視頻序列圖像中,相鄰的圖像之間具有連續性,當視頻圖像中有運動目標時,由于運動目標的運動,相鄰圖像間的像素點灰度值差別就會較大,相反,當視頻圖像中沒有運動目標時,相鄰圖像間的像素點素灰度值差別就會較小,幀差法就是利用視頻圖像的這一特性進行檢測的,它是運動目標檢測的最簡單方法。幀差法是先用相鄰兩幀做差分運算,然后做二值化處理,從而檢測出運動目標。幀差法的基本運算原理框圖如下:
圖2.1 幀差法基本原理
在二值化的差分圖像中,取值為 0 的像素點代表變化較小或是無變化的區域,表示為背景區域;取值為 255 的像素點代表變化的區域,表示為運動目標。至此,大多數的運動目標的基本形狀已經凸顯出來了。為了能夠精確的提取出運動目標,通常還需要經過形態學處理,例如,膨脹、腐蝕、開運算、閉運算等,將斷點進行連接或者將多余的部分去掉等,從而獲得更加準確的檢測結果。
幀差法優點是算法簡單、容易實現、檢測速度較快、能滿足系統對實時性的要求,而且一般相鄰兩幀的時間間隔比較短,因此對場景的光線變化不是特比敏感,對環境的自適應性較強。幀差法存在著兩個主要的缺點:幀差法的檢測結果與目標運動速度和相鄰兩幀間隔大小有關。如果運動目標速度過快且兩幀時間間隔長,可能會出現運動目標在運動方向上前后分裂并拉長,出現“雙影現象”,最壞的情況下可能發生同一個運動目標被檢測為兩個不同的目標。如果運動目標運動過慢且兩幀時間間隔小,交疊的部分在兩幀圖像進行差分時會因差值過小而被誤判為背景區域,但是事實上這部分并不是背景區域,由此會造成目標信息的丟失,我們稱之為“空洞現象”,最壞的情況下可能發生目標完全重疊而不能被檢測出來。
運動目標檢測方法研究總結
2.3 背景差分法
背景差分法又叫做背景減除法,是固定場景中目標檢測算法最長用的一種檢測算法。在背景差分法中,視頻圖像分為背景圖像和前景圖像,該方法將視頻圖像中的當前圖像與背景圖片進行比較,也就是當前幀與背景幀進行差分運算,在運算結果中像素點的灰度值變化大的區域即為運動目標,因此我們也常常認為背景差分法是幀間差分法的一種特例。
背景差分法的運算原理框圖如下:
圖2.2 背景差分法基本原理框圖
提取出運動目標區域之后,可能會存在噪聲,可以對其進行后期處理,例如膨脹、腐蝕、連通域檢測等操作,從而提取出較為準確的運動目標。背景差分法的優點是算法復雜度低,算法實現比較簡單,可以滿足系統的實時性要求,并且在運動目標檢測時一般能夠得到比較完整的特征數據。是目前最常用的一種運動目標檢測方法,尤其適用于背景固定或背景緩變的場景。該方法對外界的環境變化非常敏感,例如場景中天氣的變化,光線的改變,攝像機的顫動,樹葉的搖動等等這些外界的干擾很容易使背景點被誤判為目標點,因而影響了檢測的精度。所以該方法的難點在于建立一個符合場景需求的背景模型,而且需要有一套算法對背景模型進行更新,使背景成為實時而準確的背景。檢測效果好的背景模型往往都會比較復雜,運算量比較大,從而使得背景更新的速度減慢,無法實時地檢測出運動目標。如果背景模型更新的速度非常慢,實際上背景在不斷變化著,會 導致從背景模型得到的背景圖像不是實時的背景圖像。但是如果背景模型更新的速度過快,背景有時就會和當前圖像非常相似,背景差分法就無法檢測出運動物體。因此,建立一個合適的背景模型是背景差分法的關鍵部分,也是該算法的一個難點。第二章 經典的運動目標檢測算法 3
運動目標檢測方法研究總結
第三章 改進的運動目標檢測算法
3.1 改進的三幀差分法
三幀差分法是對兩幀差分法的改進,它可以有效的克服兩幀差分法的一些缺點,例如,克服了兩幀差分法中的“雙影現象”,但是卻不能改善“空洞現象”,而且當運動目標和背景區域這兩部分的灰度值比較接近時,三幀差分法不能準確的檢測出運動目標的輪廓.為了改善三幀差分法檢測結果的邊緣缺失的這個缺點,王霏等人嘗試將改進的Sobel算子與三幀差分法相結合,因為圖像的邊緣信息不容易受噪聲和亮度突變的影響。改進算法的關鍵是獲得一個完整的運動目標的邊緣輪廓,針對“空洞現象”,通過后期的形態學處理和連通性分析相結合的方法來改善這個問題[20]。
改進的三幀差分法的基本思想是:把三幀分成兩組,分別求兩幀差分圖,進行膨脹處理之后分別和邊緣檢測結果圖相“與”,將兩個結果進行“或”運算,得到一個初步的檢測結果。然后與三幀差分法的結果進行“或”運算,得到一個更加完整的檢測結果。最后對檢測結果進行后期處理,來達到改善“空洞現象”和去除噪聲的目的。
(a)經典的兩幀差分法
第三章 改進的運動目標檢測算法 7
(b)經典的三幀差分法
(c)改進的三幀差分法
3.2 幀間差分法與光流法結合
光流法對噪聲敏感,運動目標邊緣以外附近的像素點沒有運動,但光流值不是零,出現了“速度漂移”。使得光流法和超像素分割的分割結果雖然不錯,但是需要調節的參數空間非常大,需要花費大量的時間來進行參數的調整,導致它的實用性較差。幀間差分法簡單易行,利用當前幀與前一幀差分,當運動目標運動過快時提取出的物體往往會出現雙影而且比實際要大,當運動過慢時又往往因無法檢測出重疊部分而出現空洞。為此,我們在光流法和超像素分割方法的基礎上,融合了幀間差分法來進一步快速的縮小需要調節的參數的范圍,極大的減少了人工調節的工作量。在改進的方法中,賀麗麗等人結合光流法產生的結果和幀間差分法的結果來提取出大致的運動區域,然后利用超像素分割進行精確的提取[21]。
利用超像素分割后的結果label,對幀間差分法得到的結果s 進行進一步的提取,由于超像素分割將圖像分割成許多含有相似特征的標記圖label,首先我們需要知道s,即已經求出的前景區域中含有對應標記圖中的哪些標記。當我們確定了前景區域所包含的標記后,就需要進一步的確定這些標記在s和label 中所包含的像素個數。對于s,我們統計出其前景區域中對應于標記圖label 中每個標記i所包含的像素的個數為isn,標記圖label中每個標記i所包含的像素的個數為n,我們認為isn與n 之間的關系,將成為提取前景目標最重要的衡量標準。如果它們的比值小于某一個值T 時,則將Label中對應于標記i的像素值設為0,否則,Label中對應于i的像素值設為 1。原始圖像t 中對應于 Label中等于 0 的位置也標記為 0;因此我們得到了提取出的運動目標圖。
運動目標檢測方法研究總結
3.3 改進的背景建模算法
針對傳統混合高斯背景建模算法,由于受到算法的限制,每個像素點的模型個數是固定不變的。后來,等人提出了基于最大似然估計的像素點背景模型個數自適應選擇方法。但由于該方法人為地引入了負的先驗系數,使得在更新過程中高斯成分的權重有可能被不合理地負更新。王永忠等人在傳統混合高斯背景建模的基礎上,提出一種了自適應選擇混合高斯模型個數的策略,孫麗等在此基礎上提出了新的算法。
其算法的基本思想是對于每幀,在圖像所有像素的模型總個數固定的情況下,對于頻繁變化的背景點,當未達到像素點自身的模型個數上限且不超過圖像模型總個數的前提下,可以“借用”相對穩定的區域像素點的未用模型個數的名額,用個數不定的模型來描述該像素點可能的背景。根據模型的轉化規則,通過刪除模型、增加模型、替換模型來動態分配模型個數及模型的參數更新,并實現背景模型的生成,從而使得背景模型個數能夠隨著場景的變化動態調整。
在當前時刻,每個像素點都有自己的初始模型,這些模型根據一定的閾值條件,被劃分成背景模型和候選背景模型。背景模型和候選背景模型都有自我更新并保持自身模型的狀態。當候選背景模型滿足一定的條件后,轉化成背景模型,而背景模型和候選模型通過刪除模型的機制來達到終止模型狀態。本文的背景建模算法在不增加模型總個數的前提下,通過動態刪除模型、增加模型以及模型間的轉化三種處理機制,分配不定個數的背景模型,并自適應更新參數。
實驗表明,該算法能夠較好的處理復雜動態場景中背景的頻繁變化如樹枝葉搖曳、水面波動和噴泉,并在構建背景過程中,一定程度上克服了前景運動目標的影響。與傳統的運動目標檢測算法相比,本文算法對動態變化背景的描述能力更強,處理速度更快,符合實際場景中運動目標檢測的實時性和準確性要求[22]。
第三章 改進的運動目標檢測算法 7 12 運動目標檢測方法研究總結
第四章 總結
運動目標檢測是計算機視覺中富有挑戰性的課題之一,對其展開研究具有重要的應用價值和理論意義。本文首先總結研究了現存的三種經典的運動目標檢測算法:光流法、幀差法和背景差分法,并對分析了它們的優缺點。光流法不需要預先知道有關場景信息,可以支持攝像機的運動,但是計算復雜度高、耗時長、抗噪性能差,目前沒有較好的通用硬件支持。幀差法是將相鄰的兩幀圖像對應像素點進行相減而得到運動目標,能夠適應光照突變,運算速度快、算法簡單,但是檢測結果容易出現“空洞現象”和“雙影現象”,它適用于對實時性要求高,對檢測的目標信息要求不高的場景中。背景差分法是將當前圖像幀和建立的背景圖片相減來得到運動目標,該算法復雜度不高,可以滿足系統對實時性的要求,背景模型的建立對該方法的實現會產生至關重要的作用然后。
然后介紹了幾種改進的檢測算法,改進的三幀差分法將改進的 Sobel 邊緣檢測算子引入到三幀差分法中,因為圖像的邊緣信息不容易受噪聲和亮度突變的影響,所以將三幀差分法和改進的 Sobel 算子相結合彌補了三幀差分法不能將運動目標輪廓完整檢測出來的缺點。然后對檢測結果進行后期處理,達到改善空洞現象和去噪的目的。最后通過實驗對比經典三幀差分法和改進的三幀差分法的檢測效果,結果表明改進的三幀差分法的檢測結果準確性更高。利用光流場獲得粗略的運動區域,將光流值經過濾波,二值化和形態學處理,獲得大概的運動目標區域。利用幀間差分法將相鄰兩幀圖像中變化的區域提取出來,通過結合光流法和幀間差分法的結果,進一步將運動目標的提取鎖定在一個更加準確的范圍內。背景建模法對于每幀,在所有圖像像素模型總個數固定的情況下,對于頻繁變化的背景點,當未達到像素點自身的模型個數上限且不超過圖像模型總個數的前提下,可以借用相對穩定的區域像素點的未用模型個數的名額,用個數不定的模型來描述該像素點可能的背景,從而實現背景模型的動態分配及更新。并基于像素的時域信息,用不包含該像素點在內的小鄰域內所有像素點的背景模型與當前點進行匹配,判斷其是否為前景點,以此消除動態背景干擾的影響。這些算法都將為今后的研究提供參考和依據。第四章 總結 11 14 運動目標檢測方法研究總結
參考文獻:
[1] 岡薩雷斯.數字圖像處理(第二版)[M].北京:電子工業出版社, 2003:1-115.[2] Stauffer C, Grimson W E L.Learning Patterns of Activity Using Real-time Tracking [J].IEEE Transactions on Pattern Analysis & Machine Intelligence, 2000, 22(8):747-757.[3] H.Akaike.A New Look at the Statistical Model Identification.IEEE Transaction on Automatic Control, 1974, 19(6):716-723.[4] 朱宏.基于視頻序列的運動目標檢測與跟蹤技術研究[D].西南交通大學碩士學位文,2008:1-5.[5] B K P Horn, B G Schunck.Determing Optical Flow [M].Artificial Intelligence, 1981:185-203.[6] Meyer D.Model Based Extraction of Articulated Objects in Image Sequences for Gait Analysis[C].Proe IEEE International Conference on Image Processing, Santa Barbara, California, 1997:78-81.[7] J L Barren, D J Fleet, S S Beauchemin.Performance of Optical Flow Techinques, Computer Vision, 2004, 12(1):43-77.[8] Roland M, Michael W.A Noise Robust Method for Shape Estimation of Moving Objects in Video Sequences Considering a Moving Camera [M].In Signal Processing, 1998:203-217.[9] 甘明剛, 陳杰, 劉勁.一種基于三幀差分和邊緣信息的運動目標檢測方法[J].電子與信息學報, 2010:894-897.[10] 郝毫剛, 陳家琪.基于五幀差分和背景差分的運動目標檢測算法[J].計算機工程, 2012, 38(4): 146-148.[11] Wren C.Real-Time Traeking of the Human Body [J].IEEE Transactions on Pattern Analysis and Machine Intelligenee, 1997, 19(7):780-785.[12] Stauffer C, Grimson E.Learning Patterns of Activity Using Real-Time Tracking [C].In IEEE [13] 左軍毅, 潘泉.基于模型切換的自適應背景建模方法[J].自動化學報, 2007, 5(33):467-472.[14] Transactions on Pattern Recognition and Machine Intelligence(TPAMI), 2000, 22(8):747-757.[15] Xuming Zhang.Impulse Noise Removal Using Directional Difference Based Noise Detector 參考文獻 13
and Adaptive Weighted Mean Filter [J].Signal Processing Letters, IEEE, 2009, 16(4):295-298.[16] Gao P, Sun X, Wang W.Moving Object Detection based on Kirsch Operator Combined with Optical Flow[C].Image Analysis and Signal Processing(IASP), 2010 International Conference on.IEEE, 2010: 620-624.[17] Wixson L.Detecting Salient Motion by Accumulating Directionally Consistent Flow[J].IEEE Trans, Pattern Analysis and Machine Intelligence, 2000(22):774-780.[18] Seki M, Fujiwara H, Sumi K.A Robust Background Subtraction Method for Changing background [J].Proceeding of IEEE workship on Application of Computer vision, 2000:207-213.[19] Qinghua Ji, Suping Yu.Motion Object Detection Based on Adaptive Mixture Gaussian Model and Four-frame Subtraction [J].International Conference on Computational and Information Sciences, 2013:1202-1205.[20] 王霏.基于視頻的運動目標檢測算法研究[D].吉林大學, 2014.[21] 賀麗麗.視頻序列中運動目標提取方法研究[D].西安電子科技大學, 2014.[22] 孫麗.基于背景建模的運動目標檢測算法研究[D].東北大學, 2011
第五篇:“目標檢測”——基于目標的教學設計
浙江臺州市黃巖區院橋鎮鎮北小學(318020)阮林萍目標檢測基于目標的教學設計(以下簡稱目標檢測),首先要明確教學目標,然后再確定檢測的內容和方式,即教學評價,最后設計教學活動。目標檢測的核心是關注教學目標,特征是教學評價先于教學設計。本文主要討論三個話題:什么是目標檢測?為什么提出目標檢測?如何實施目標檢測?
一、概念界定:什么是目標檢測目標檢測是基于目標的教學設計,它并非全新的教學理念,崔允漷教授的《基于課程標準的教學》為本課題研究提供了很好的理論支持和啟發。基于課程課程標準的教學不是一種教學方法,而是一種理念?;跇藴实慕虒W需要教師在對標準深刻理解的基礎上,把握對學生的總體期望,將標準轉化為年級目標,再根據學生特點和教學情境具體化為每一堂課的教學目標。目標檢測是將教學目標、教學評價和教學活動融為一體的教學設計思路。這種教學設計,第一步是明確課時
浙江臺州市黃巖區院橋鎮鎮北小學(318020)阮林萍目標檢測基于目標的教學設計(以下簡稱目標檢測),首先要明確教學目標,然后再確定檢測的內容和方式,即教學評價,最后設計教學活動。