h1

PAL 全景相机图像校正

2011/02/23

The camera using panoramic annular lens (PAL) can capture the surrounding scene in a view of 360° without any scanning component. Due to severe distortions, the image formed by PAL must be unwrapped into a perspective-view image in order to get consistency with the human’s visual custom. However the unfilled pixels would probably exist after unwrapping as a result of the non-uniform resolution in the PAL image, hence the interpolation should be employed in the phase of the forward projection unwrapping. We also evaluated the performance of several interpolation techniques for unwrapping the PAL image on a series of frequency-patterned images as a simulation by using three image quality indexes: MSE, SSIM and S-CIELAB. The experiment result revealed that those interpolation methods had better capability for the low frequent PAL images. The Bicubic, Ferguson and Newton interpolations performed relatively better at higher frequencies, while Bilinear and Bezier could achieve better result at lower frequency. Besides, the Nearest method had poorest performance in general and the Ferguson interpolation was excellent in both high and low frequencies.

傳統的監控系統若要實現全方位的監控通常採用電機帶動攝像頭的旋轉來完成,這種方式的缺點是不能同時監控360度區域,實時性能差。而多點安裝攝像頭成本則較高,也不能瞬間獲得所有視場。在這樣的需求下,基於全景鏡頭的視頻監控系統應運而生,該應用使用一個全景環形透鏡(PAL)相機即可獲得監控空間全方位的視場。

全景攝像頭如圖所示,全景環形透鏡(PAL),在這種成像機制中,沒有掃描器件存在,能夠實現實時360度成像,可以取代常規的樓層和辦公室的監控攝像機,獲得360度的大視場。系統結構小巧,尺寸和普通的攝像機相當,圖像畸變小,恢復方法簡單,可廣泛安裝在樓層和辦公室,取代傳統的複雜的帶有掃描和轉動的監控設備。
全景攝像頭主要功能是採集全景圖像,並且輸入標準的PAL制式模擬視頻。採集後的圖像如圖所示。展開後的圖像如圖(在我們開發的軟件中展開)。

全景圖像的展開我們開發了不同的方法,以適用於不同的應用場合。從運算時間上、空間頻率上以及實際效果上我們做了詳細分析,並採用了常見的MSE和SSIM指數評價展開方法的圖像保真度,如下圖所示,以下橫坐標是我們採用的不同展開算法,不同的曲線對應不同的圖像空間頻率,MSE越小而SSIM越接近1則圖像保真度越高,實際使用中可根據具體場景分析最適合展開方法。

展開後的全景圖像亦可應用於全景監控等領域,此時圖像與常規圖像一直,可結合特徵提取、機器學習等實現智能視頻監控(如本站中提到的SmartCam),如入侵預警、人形檢測、人臉探測等,亦可用於車載視覺實現駕駛員疲勞度監控,車牌識別,行人探測,障礙物探測等等。
Submitted by Admin on Thu, 06/10/2010 – 21:55. Applications工業應用

http://www.dvzju.com/Page_PAL

發表留言