車週監控系統之攝影機校正及影像接合

車週有許多盲點區域會造成駕駛者難以判斷車週情況以導致交通事故的發生，威脅到駕駛者與行人的安全。為了提高行車安全性，輔助駕駛系統提供了車週狀況的警告信號及視覺提醒。然而，現今大部分的輔助駕駛系統只能提供不夠寫實的車週影像。因此我們研發了一個寫實且即時的車週監控系統。而這個系統中，我們用安裝在車上的四個魚眼攝影機以及一個深度攝影機來擷取車週影像資訊及深度資訊，並且整合所有的資訊以提供駕駛者一個可轉換視角的車週影像。這次的研究分成三個部分，攝影機校正，影像接合以及深度調變混合投影模型。我們提出了一個精準的攝影機校正方法先將原始扭曲的影像還原成針孔投影無扭曲的影像，再利用階層式 RANSAC 將所有的攝影機統一到同一個世界座標上。接著進行影像的接合以整合所有的影像資訊。原理是要使得影像的重疊區域有個平滑的亮度變化，因此提出了一套亮度均化以及影像融合的程序。而為了架構出整個可變視角的監控系統並且消除投影在地面的影像扭曲，我們仍然提供一個立體模型以投影接合完的影像。簡而言之，我們研發了一套即時的車週影像系統，可以提供駕駛者選擇一個最適合目前駕駛情況的視角來獲取輔助駕駛的影像資訊。

關鍵字

監控系統；魚眼攝影機校正；影像接合

並列摘要

Vehicles usually have blind spots that make driver difﬁcult to judge the real situation of the surrounding environment and eventually threaten the safety of drivers and pedestrian. For the driving safety, driving assistance systems provide warning signals or visual cues of sur- rounding situation. Most of existing driving assistance system can only provide unrealistic vehicle surrounding images. In this work, we develop a vehicle surrounding monitoring system with realistic perception in real time. In our system, four ﬁsheye cameras and one stereo camera are mounted on the four sides of the vehicle to capture the videos of the vehicle surrounding views. This system integrates all the image information and provides user some vehicle surrounding images in changeable viewpoint. The proposed procedure to construct this system are divided to three part: camera cal- ibration, images stitching, and the depth adaptive hybrid model. We propose a novel and accurate ﬁsheye camera calibration method to dewarp the captured images into perspec- tive projection ones. These images are then spatially transformed to a ground plane with their own homography matrices computed by a hierarchical RANSAC procedure. Then we propose an image stitching technique to integrate all the image information. To smooth the brightness change in the overlapping region, we propose new method for brightness uniﬁcation and image blending. For variable-viewpoint monitoring, the stitched image on the ground plan is projected to a 3-D hybrid projection model which can eliminate the dis- tortion caused by the homography process and makes the ﬁnal rendered view look more realistic. Moreover, we have developed an online system in which the driver can choose the appropriate view to look around the surrounding environment of the vehicle.

並列關鍵字

Surrounding monitoring system ； Fisheye camera calibration ； Image stitching

參考文獻

[1] Driving assistance system in Audi A6. Audi.

[2] A threshold selection method from gray-level histograms. Systems, Man and Cyber-

[4] C. Allene, J.-P. Pons, and R. Keriven. Seamless image-based texture atlases using

[5] M. Aly. Real time detection of lane markers in urban streets. Intelligent Vehicles

Symposium, 2008 IEEE, pages 7–12, June 2008.

國際替代計量

車週監控系統之攝影機校正及影像接合

全文下載

主題瀏覽