11320E500100 工學院英文書報討論 Seminar

主題 TOPIC
▸　基於單眼立體視覺之三維街景重建 Three-Dimensional Street-View Reconstruction with Monocular Stereo Vision

❝　隨著智慧車輛研究的廣泛發展，光偵測和測距(light detection and ranging, LiDAR)被廣泛地應用於三維(three-dimensional, 3D)街景的偵測和重建。然而，為了滿足物體辨識的需要，其需要依賴影像資訊的輔助。因此，越來越多的研究利用立體視覺技術來解決3D空間偵測或物件辨識的需求。獲取每個影像對的準確運動資訊，是對於透過移動單一相機捕獲的連續影像，來進行3D街景影像拼貼的一個困難挑戰。本研究提出了一種使用單一攝影機實現2D到3D 影像轉換的3D街景重建演算法。此演算法能夠準確估計相鄰影像對的相對基線長度，從而準確地將2D影像轉換為3D影像，並進一步完成多組3D影像的拼貼。此外，所提出的演算法還可以過濾掉3D估計誤差、移動物體和物體拖曳問題。在實驗結果中，對所提出的演算法和三種視覺同步定位與建圖(visual simultaneous localization and mapping, VSLAM)相關演算法進行了分析，並與測試影像進行了比較。結果證實了所提出的演算法在準確重建3D街景方面的優越性。

❝　With the extensive developments of intelligent vehicle research, LiDAR (light detection and ranging) is widely used in the detection and reconstruction of three-dimensional (3D) street view. However, in order to meet the needs of object recognition, it is necessary to rely on the assistance of image information. Therefore, more and more researches have used stereo vision technologies to solve the needs of 3D space detection or object recognition. A difficult challenge for 3D street view image collage of continuous images captured by moving a single camera is to obtain accurate motion information for each image pair. This research proposes a 3D street-view reconstruction algorithm using a single camera to achieve 2D-to-3D image conversion. The proposed algorithm can accurately estimate the relative baseline length of adjacent image pairs, thereby accurately converting 2D images into 3D images and further completing the collage of multiple sets of 3D images. In addition, the proposed algorithm can also filter out 3D estimation errors, moving objects and object dragging issues. In the experimental results, the proposed algorithm and three VSLAM (visual simultaneous localization and mapping) related algorithms are analyzed and compared with test images. Results confirm the superiority of the proposed algorithm for the accurate reconstruction of 3D street views. 　❞

講者 SPEAKER
▸　徐勝均教授 Prof. Sendren Sheng-Dong XU
▸　國立清華大學動力機械工程學系 Department of Power Mechanical Engineering, National Tsing Hua University

時間 TIME
▸　2025/05/27 (TUE) 13:20 ~ 15:10

地點 VENUE
▸　工程一館201教室 Classroom 201, Engineering Building 1