運用新式立體視覺演算法於面型及線型攝影機三維模型重建技術之研究

A Study on Three-Dimensional Model Reconstruction Technology for Area and Line Cameras Using a Novel Stereovision Algorithm

陳建宏(CHIEN-HUNG CHEN)

指導教授 ： 陳冠宇

中原大學/工學院/機械工程研究所/博士(2017年)

https://doi.org/10.6840/cycu201700045

摘要

自動化光學檢測(automatic optical inspiration, AOI)為整合影像處理、光學設備及自動化技術，對產品進行非接觸檢測，提供高速度、高效率、高穩定、高重複的線上檢驗，是現代自動化生產不可或缺的一環。然而現今AOI檢測技術大多為二維平面影像的應用，可預期的是：當電腦硬體計算效率越來越好，對待測物的檢驗將不再僅限於平面位置，會需要對待測物的曲面深度有全面的感知能力，因此，基於三維影像的AOI系統必會愈趨普及。此外，面型攝影機普遍應用於現今的AOI系統進行影像擷取與分析，但隨著工件尺寸及精度等級一再向下突破，AOI系統的精度及處理速度也必須隨之升級，此時面型攝影機的解析度及取像速度即可能無法滿足生產線的要求，而線型攝影機因其高解析度及高速取像的特性相較於面掃描攝影機更具優勢，成為提升AOI系統的一種有效解決方案。因此，本文提出基於新式立體視覺演算法的線型攝影機之三維影像重建技術，以獲得更精確的三維影像模型。首先，本文分別建置由面型及線型攝影機構成的立體視覺影像擷取系統，再進行影像校正，最後結合新式立體視覺演算法及三種深度視差法：半全域匹配、區域局部匹配及影像切割演算法，分別進行三維影像模型的重建，藉由分析比較，找出最佳的三維影像模型重建方案。根據實驗結果，本文的主要成果有三：(1)以半全域匹配演算法為核心，利用前後各一組立體視覺攝影機模組進行拍攝，藉此重建目標物360度的三維模型；(2)將半全域匹配演算法得到的深度模型結合新式立體視覺演算法並對非均勻離散層級與仿射幾何空間進行誤差修正，得出更精確的深度數值；(3)利用線型攝影機模擬生產線中對目標物連續進行影像擷取，完成重建其三維模型。

關鍵字

立體視覺 ； 機器視覺 ； 半全域匹配 ； 三維重建 ； 線型掃描攝影機

並列摘要

In order to integration the image processing, optical equipment and automation technology on automated optical inspection. Provide high speed, high efficiency, high stability, and high repeatability is indispensable part of automated production on non-contact detection. In recent years, most AOI detection technology is about the application of 2D planar imaging. It is expected that: When the computer hardware computing efficiency is getting better, detect of the object will no longer be limited on plane, we need to measure the surface depth of object comprehensively, Therefore, the AOI system based on the three-dimensional detection will become popular. Area cameras are used in today's AOI system for image capture and analysis widely. However, the smaller workpiece size and higher precision grade are required, accuracy and processing speed of AOI system also need to be upgraded, resolution and capture speed on area cameras may not fulfill the requirements of the production line. To promote the ability on AOI system, using line camera could be an effective solution because of its high-resolution and high-speed Image capture. Therefore, this paper proposes a three-dimensional image reconstruction technique for linear cameras based on a novel stereovision algorithm to obtain more accurate three-dimensional image model. First, the paper construct the surface type and line type camera composed of stereo vision image capture system, respectively. Then correct left and right image. Finally, combined with novel stereovision algorithm and compare three kinds of depth matching method: Semi-Global Matching algorithm, Local Feature Matching, Grapth Cuts algorithm. Reconstruction three-dimensional model, respectively. Analyzing and compare the result, then find the best three-dimensional image reconstruction model. According to the experimental results, the results of this paper are as follows: The first we reconstruct the 360° three-dimensional model by two pairs CCD cameras shooting front and behind the target, and use Semi-Global Matching (SGM) algorithm compute the depth of target. The second combine Semi-Global Matching (SGM) algorithm with novel stereovision algorithms and use non-uniform spacing of discrete depth level error correction and affine space warping error correction to fix the depth of target. The third construct a detection system to simulate the production line and reconstruct the 3D model of target by two line-CCD camera

並列關鍵字

stereovision ； machine vision ； semi-global matching (SGM) ； three-dimensional reconstruction ； line-CCD camera

參考文獻

[22] 鍾宏英，多自主行動機器人無線感測網路室內定位系統之研究，碩士論文，中原大學機械工程學系，2014。

[32] 郭瑞濠，立體視覺多目標追蹤系統之發展，碩士論文，中原大學機械工程學系，2009。

[26] 張嘉麟，模糊邏輯控制可搭乘電梯之立體視覺伺服輪型機器人，碩士論文，中原大學機械工程學系，2005。

[28] 劉易軒，視覺伺服擲球機械手臂之研究，碩士論文，中原大學機械工程學系，2010。

[30] 曾彥博，立體視覺伺服雙目標追蹤系統之研究，碩士論文，中原大學機械工程學系，2009。