查詢結果分析
來源資料
頁籤選單縮合
題 名 | 單張影像之三維物體識別與定位的快速演算法則=Polyhedral Recognition and Localization with Fast Library Search from a Monocular Image |
---|---|
作 者 | 何信瑩; 黃政池; | 書刊名 | 逢甲學報 |
卷 期 | 31 1997.06[民86.06] |
頁 次 | 頁151-170 |
分類號 | 312.1 |
關鍵詞 | 三維物體識別; 物體定位; 物體表示法; 視覺成員; 不變特徵量; 3-d object recognition; Pose determination; Object representation; Library view; Invariant feature; |
語 文 | 中文(Chinese) |
中文摘要 | 本論文提出一套資料庫編譯法則,配合一套有效率的資料庫搜尋演算法,使得由單 張黑白影像中所取得之多面體的二維資訊可以迅速從大量資料庫中比對出正確的物體和其方位 。我們提出一套描述二維影像特徵與模型物體特徵之法則來組成單一物體表示法,它可有效解 決因多個物體互相重疊或自身遮掩所產生的遮蔽問題。將每一物體之所有可見完整面的拓蹼特 徵加以編譯,找出最佳的搜尋鍵,使得在比對時,未知物能以搜尋鍵值快速找到候選物再作細 部比對。最後,再針對被局部遮掩的平面部份作檢驗,並決定其方位。 |
英文摘要 | In this paper an efficient polyhedral object recognition and localization method is presented. A new multiview representation of an object which is extracted from a 2-D image and stored in a database is presented. It can handle the occlusion problems due to the object overlapping or selfocclusion. Some topological features of sensed objects are used as the search key to select the small set of candidate object models and possible views. Furthermore, perspective projections of the possible model views are used to verify the 2-D line features of the sensed object. If the correct classification of objects is made, the object positions can be determined. |
本系統中英文摘要資訊取自各篇刊載內容。