應用影像光流法於土石流表面顆粒之追蹤

Image Optical Flow Applied to Trace Particle on Debris Flow Surface

黃煒翔(Wei-Hsiang Huang)

指導教授 ： 劉格非

國立臺灣大學/工學院/土木工程學研究所/碩士(2019年)

https://doi.org/10.6342/NTU201903610

摘要

本研究建立於影像光流法之基礎上，將現實三維環境之運動場，投影至圖像二維平面，利用圖像序列中像素亮度在時間域上的變化，以及相鄰影像之間的相關性來找到上一幀跟當前幀之間存在的對應關係，從而計算出相鄰影像之間物體的運動信息的一種方法。由於光流法擁有許多不同分支，本研究主要為Farneback光流法(Farneback optical flow )進行改良及優化，主要概念為二次擬合像素點周圍之亮度趨勢分布，在時間序列上，將追蹤不同圖像有著相同趨勢分布的像素點，進而求取像素點之運動信息
為了從圖像內得到具有物體代表性之資訊，透過像素點之間的亮度梯度變化，明確找出物體邊界，試著追蹤這些邊界上的像素點，從而計算光流的變化，獲取二維圖像運動場。本研究目的為應用於土石流事件，將進行室內小型實驗，模擬土石流中有著不同粒徑顆粒情況；由於二維影像僅投影三維物體表面，故透過光流法之計算，得表面顆粒之運動信息，最後將本研究方法，真實應用於現場土石流事件，並整理與分析。

關鍵字

影像分析 ； 表面流速 ； 亮度梯度 ； 影像光流法 ； 二次函數擬合

並列摘要

Based on the image optical flow method, this study projects the motion field of the realistic three-dimensional environment to the two-dimensional image of the image, using the variation of the pixel brightness in the time domain of the image sequence, and the correlation between adjacent images. A method for finding the correspondence between the previous frame and the current frame to calculate the motion information of the object between adjacent images. Since the optical flow method has many different branches, this study mainly improves and optimizes the Farneback optical flow method. The main concept is the quadratic fitting of the brightness trend distribution around the pixel points. In the time series, the tracking will be different. The image has pixels with the same trend distribution, and then the motion information of the pixel is obtained.
In order to obtain the representative information of the object from the image, the boundary of the object is clearly found through the change of the brightness gradient between the pixels, and the pixel points on the boundary are tried to calculate the change of the optical flow to obtain the two-dimensional image. Like a sports field. The purpose of this study is to apply to small-scale indoor experiments to simulate the case of different particle sizes in the earth-rock flow. Since the two-dimensional image only projects the surface of the three-dimensional object, the motion information of the surface particles is obtained by the calculation of the optical flow method. Finally, the research method is applied to the on-site earth and rock flow events, and is sorted and analyzed.

並列關鍵字

image processing ； surface velocity ； intensity gradient ； optical flow ； quadratic function fitting

參考文獻

[1] Gunnar Farneback,(2003). “Two-Frame Motion Estimation Based on Polynomial Expansion”

[2] Gunnar Farneback,(2002). “Polynomial Expansion for Orientation and Motion Estimation”, Link¨oping Studies in Science and Technology. Dissertations No. 790,

[3] Lucas B and Kanade T, (1984).“An Iterative Image Registration Technique with an Application to Stereo Vision” Proc. ,Of 7th International Joint Conference on Artificial Intelligence (IJCAI)

[4] Bruce David. Lucas,(1984).“Generalized Image Matching by the Method of Differences” doctoral dissertation, tech. report , Robotics Institute, Carnegie Mellon University, July

[5] Chia-Ming Wangi, Kuo-Chin Fani, Cheng-Tzu Wang .(2008). “Estimating Optical Flow by Integrating Multi-Frame Information” , Journal of Information Science and Enginering 24, 1719-1731