- 學位論文
改善Canny邊緣檢測法之多方向新型雜訊濾波器的設計並實現於無人機視覺檢測分析橋樑裂縫
A New Multi-Directional Median Filter Design to Improve Canny Edge Detector for Bridge Crack Analysis Realized via UAV Vision System
摘要
以人類眼睛而言視覺佔人體60%、聽覺佔20%、其餘佔20%，往往影像訊息都是由彩色訊息轉為灰階訊息再由灰階訊息進行著手，若在複雜背景的環境當中，不必要的訊息太多，而會有無法偵測裂縫的情況。邊緣檢測技術發展已有三十多年的歷史，其中傳統Canny演算法是利用影像梯度以及搭配雙閾值的方式來抓取邊緣的技術，高斯濾波器雖然能夠抑制雜訊但同時削弱邊緣以及細節的訊息。因此，本論文提出改善Canny演算法為基礎應用於橋樑裂縫檢測增強邊緣細節訊息，必須考慮到遮罩視窗內每個像素值是否受到雜訊所影響而增加了十二個方向，為了確保像素值不會產生誤判情況增加範圍定義來提升偵測的結果，對於是否受到雜訊所干擾的像素值增加模糊範圍概念定來確保像素值的好壞，提高邊緣的完整性和連續性擴展Sobel演算法為四個方向，經過實驗結果可以得到，即使在複雜的環境背景當中並將圖像資料轉換成我們人眼更容易理解的範圍都能夠有效保留裂縫邊緣的訊息。
並列摘要
In terms of the human eye, vision accounts for 60%, hearing accounts for 20%, the rest accounts for 20%, often image messages are changed from color messages to gray-scale messages and then by gray-scale messages, if in a complex background environment, unnecessary information is too much, and there will be unable to detect cracks. Edge detection technology has been developed for more than three decades, in which the traditional Canny algorithm is the use of image gradients and a combination of double threshold scrimmage edge technology, Gaussian filters can suppress noise but also weaken the edge and detail of the message. Therefore, this paper proposes to improve the Canny algorithm as the basis for the bridge crack detection enhanced edge detail information, must take into account whether each pixel value in the shield window is affected by noise and increased by twelve directions, in order to ensure that the pixel value does not create a misjudgment increase range definition to enhance the detection results. For whether or not the pixel value is disturbed by noise increases the fuzzy range concept to ensure the quality of the pixel value, improve the integrity of the edge and continuous extension Sobel algorithm for four directions, the experimental results can be obtained, Even in complex environmental contexts and converting image data into a range that is easier understood by our human eye, it effectively preserves information about the edge of the crack.