機車偵測演算法之研究

淡江大學機構典藏 > 商管學院 > 運輸管理學系暨研究所 > 學位論文 > Item 987654321/33827

jsp.display-item.identifier=請使用永久網址來引用或連結此文件: https://tkuir.lib.tku.edu.tw/dspace/handle/987654321/33827

题名:	機車偵測演算法之研究
其它题名:	The study of the motorcycle detection algorithm
作者:	莊劍嵐;Chuang, Jian-lan
贡献者:	淡江大學運輸管理學系碩士班范俊海;Fan, Chun-hai
关键词:	機車偵測;遮蔽偵測;區塊切割;車輛種類;交通流量;Motorcycle Detection;Occlusion Detection;Block Segmentation;Vehicle Classification;Traffic flow
日期:	2005
上传时间:	2010-01-11 04:34:54 (UTC+8)
摘要:	近年來由於伴隨著ITS的蓬勃發展，影像處理技術也日趨受到重視，雖然在技術方面有大幅度的進步，但是仍未達到成熟的境界。而綜觀國內外學者之研究，多著重於小型車之辨識與偵測，對於機車之辨識偵測皆未有專門的探討，這除了由於國外交通情況不同，也與機車影像本身的特性有很大的關連。本研究首先針對機車影像之特性做一番探討：1、機車動線紊亂; 2、機車外形及騎士影像顯現複雜; 3、遮蔽問題嚴重。並針對其特性逐步構建機車偵測演算法。本研究所發展之「機車偵測演算法」，有別於過去研究對於車輛以「輛」為單位的處理方式，本研究提出以進入偵測區到離去偵測區塊之亮塊變化的「串」為單位來進行處理，更能有效的針對目標車輛進行分析、切割等動作。整個演算法可分為三大部份，「前處理」、「亮塊關係處理」及「區塊關係處理」。「前處理」主要是利用背景相減、二值化等方法去標記出亮塊位置，並加以記錄。「亮塊關係處理」依靠了連續兩張影像間亮塊重心的位置及彼此間的交集關係，去判定是否為同一亮塊，並以亮塊變化的「串」為單位，對各串進行處理。最後以亮塊關係為基準，進行「區塊間關係處理」，例用機車富涵紋理之結果，進行區塊匹配、區塊切割等工作，來將個別車輛分開。本研究所析出之交通參數包括車流量與車速，經過實例驗證之後小型車與機車之辨識正確率，分別為87.61%及93.16 %；而遮蔽情況下小車平均有85.89％的辨識正確率，而機車平均有89.59％的辨識正確率。 In recent years, with the rapid development of ITS, the image processing technique is being paid more attention day by day. However, the traffic situation differs from Taiwan and foreign countries, so does motorcycle image characteristics itself. As results of that, the passing relevant researches were more focused on detecting car image, not motorcycle image. In this research, we will analyze the characteristics of the motorcycle image：1.The motorcycle moves confusedly, 2.Motorcycle appearance and driver’s image appear complicatedly, 3.Serious occlusion, in order to develop the “motorcycle detection algorithm”. ”Motorcycle detection algorithm” is different from the studies in past researches taking “one vehicle” as unit of detection, it takes “bunch” as a unit of detection. A “bunch” means the connecting blobs enter and leave detection zone. “Motorcycle detection algorithm” can be divided into three parts, “image preprocessing”、”blob relation processing” and “block relation processing”. First, “image preprocessing” mainly utilizes the background subtraction and threshold to label and record blob’s positions. Secondly, “blob relation processing” is depended on the blob of the center-of-gravity position and common factor relations between each two images in succession, going to judge whether or not it is the same blob. Then we take “bunch” as a unit of detection, dealing with every “bunch”. Finally, we utilize the “block relation processing” to carry on block to block segmentation match etc, to separate the specific vehicle. In our study, we have successfully extracted the traffic parameters which are included vehicle classification and traffic flow. After our experimental analysis, the rate of accurately identifying small vehicles is 87.61 %, and 93.16 % for motorcycles. Under the occlusion situations, the accurate rate of identifying small vehicle is 85.89 %, and 89.59 % for motorcycles.
显示于类别:	[運輸管理學系暨研究所] 學位論文

文件中的档案:

档案	大小	格式	浏览次数
	0Kb	Unknown	287	检视/开启

在機構典藏中所有的数据项都受到原著作权保护.

TAIR相关文章

数据加载中.....