結合深度學習與點雲分析之噴槍姿態辨識系統研究;Development of a Spray Gun Pose Recognition System Integrating Deep Learning and Point Cloud Analysis

NCU Institutional Repository > 工學院 > 機械工程研究所 > 博碩士論文 > Item 987654321/96438

jsp.display-item.identifier=請使用永久網址來引用或連結此文件: http://ir.lib.ncu.edu.tw/handle/987654321/96438

题名:	結合深度學習與點雲分析之噴槍姿態辨識系統研究;Development of a Spray Gun Pose Recognition System Integrating Deep Learning and Point Cloud Analysis
作者:	黃凱鴻;Huang, Kai-Hung
贡献者:	機械工程學系
关键词:	姿態辨識;人工智慧;Mask R-CNN;YOLOv8
日期:	2025-01-17
上传时间:	2025-04-09 18:29:17 (UTC+8)
出版者:	國立中央大學
摘要:	本研究目的在於開發一套基於深度學習與點雲分析的汽車噴漆噴槍姿態辨識系統，以解決傳統人工檢驗在噴漆作業效率低下與準確性不足的問題。此系統透過結合深度學習模型與 ZED 2 深度攝影機，以實現噴槍姿態與被噴物件位置辨識。實驗架構採用鋁擠型支架固定噴槍與目標物板子，並設置 LED 輔助光源以降低環境干擾。在研究方法中，首先對深度攝影機捕捉的影像進行前處理，利用深度學習模型進行語意分割，提取噴槍及目標物的特徵點與位置資訊。選擇的深度學習模型包括 YOLOv8 與 Mask R-CNN。接著應用點雲濾波技術去除點雲數據中的背景雜訊與多餘資訊，從而精確提取場景中目標的三維座標數據。隨後，根據提取的三維數據計算出噴槍與目標物之間的相對距離與角度。此過程結合了語意分割、點雲處理與幾何計算等多項技術，確保數據的準確性與可靠性。研究結果顯示，YOLOv8 在即時處理效能方面表現出色，而 Mask R-CNN 則在精度與細節分割能力上更具優勢。實驗結果進一步顯示，系統對角度與距離的測量誤差均在可接受範圍內，驗證了系統模型的穩定性與準確性。綜合模型的表現，本研究證實結合深度學習與點雲分析技術能有效提升噴漆作業中辨識噴槍姿態的精度與效率，展現出其在工業應用中的潛力。 ;This study aims to develop a spray gun pose recognition system for automotive painting, leveraging deep learning and point cloud analysis to address the inefficiencies and inaccuracies associated with traditional manual inspection methods. By integrating a deep learning model and a ZED 2 depth camera, the system facilitates the identification of spray gun pose and the position of target object. The experimental setup employs aluminum extrusion supports to fix the spray gun and target plate while incorporating LED auxiliary lighting to mitigate environmental influence. In the proposed methodology, images captured by the depth camera undergo preprocessing, followed by semantic segmentation using deep learning models to extract key features and positional information of the spray gun and target object. The selected deep learning models include YOLOv8 and Mask R-CNN. Subsequently, point cloud filtering techniques are applied to remove background noise and extraneous data, enabling precise extraction of the three-dimensional coordinates of the targets. Based on the extracted 3D data, the relative distance and angle between the spray gun and the target object are calculated. This process combines multiple techniques, including semantic segmentation, point cloud processing, and geometric computation, to ensure data accuracy and reliability. The experimental results indicate that YOLOv8 excels in real-time processing performance, while Mask R-CNN demonstrates superior accuracy and detailed segmentation capabilities. Additionally, the system’s measurements of angles and distances show acceptable error margins, verifying the stability and accuracy of the proposed model. Overall, the findings confirm that integrating deep learning with point cloud analysis can significantly enhance the precision and efficiency of spray gun pose recognition in painting operation, highlighting its potential for industrial applications.
显示于类别:	[機械工程研究所] 博碩士論文

文件中的档案:

档案	描述	大小	格式	浏览次数
index.html		0Kb	HTML	15	检视/开启

在NCUIR中所有的数据项都受到原著作权保护.

社群 sharing

数据加载中.....