삼각 메시로 구성된 플랜트 캐드 데이터로부터의 프리미티브 인식 (Primitive Recognition from Triangular Mesh of Plant CAD Data)

서지정보

· 발행기관 : 한국CDE학회
· 수록지 정보 : 한국CDE학회 논문집 / 25권 / 1호 / 56 ~ 66페이지
· 저자명 : 김영두, 응웬콩홍퐁, 최영

초록

As plant construction grows on a large scale, plant-specific software products are developed, with large scale design, efficient compression, and visualization capabilities. Since such visualization products are expensive, in most construction sites, inexpensive visualization software is utilized with data converted to a neutral format. Many studies have proposed a method of extracting primitive shapes from plant design data to reduce data weight. Although primitive recognition methods for various shapes such as cylinders, cubes, planes, spheres, etc. have been proposed, there is a lack of research on the classification of torus, which serves as an elbow connecting pipes. In this paper, we propose a primitive recognition method which recognizes cylinders, tori, planes, cubes, spheres from unorganized triangular mesh data of plant CAD data.
Convolutional neural network, the Gaussian sphere, and oriented bounding box are utilized to classify each primitive. Recognized primitive shapes and unrecognized mesh are exported in a lightweight format. Experiments showed high primitive recognition rates and data compression rates.

영어초록

As plant construction grows on a large scale, plant-specific software products are developed, with large scale design, efficient compression, and visualization capabilities. Since such visualization products are expensive, in most construction sites, inexpensive visualization software is utilized with data converted to a neutral format. Many studies have proposed a method of extracting primitive shapes from plant design data to reduce data weight. Although primitive recognition methods for various shapes such as cylinders, cubes, planes, spheres, etc. have been proposed, there is a lack of research on the classification of torus, which serves as an elbow connecting pipes. In this paper, we propose a primitive recognition method which recognizes cylinders, tori, planes, cubes, spheres from unorganized triangular mesh data of plant CAD data.
Convolutional neural network, the Gaussian sphere, and oriented bounding box are utilized to classify each primitive. Recognized primitive shapes and unrecognized mesh are exported in a lightweight format. Experiments showed high primitive recognition rates and data compression rates.

참고자료

· 없음