隨著科技的進步，民眾對行車安全的要求也愈來愈高。在停車或倒車的過程中駕駛通常需要轉頭或是利用後照鏡來看車輛後方與側後方區域。但車體的結構與後照鏡的角度會影響後視視野而導致有部份盲點且同時難以注意車輛週遭其餘區域的情況，使得駕駛無法全面顧及車輛週遭環境。為提高停車時的便利性與安全性，我們提出一套環場鳥瞰停車輔助系統，駕駛者無需轉頭即可由合成的環場鳥瞰影像了解週遭是否有障礙物及障礙物與車輛的相對位置，可大幅提升行車的安全。此系統主要工作包含了相機參數校正、影像扭曲校正、影像暗角補償、鳥瞰轉換、鳥瞰影像對位、亮度調整、及色彩混合等步驟。 本系統在車輛四周架設四個廣角相機以拍攝車輛週遭影像，但廣角相機雖然有寬廣的視角但也有嚴重的成像扭曲與暗角效應。我們利用基於平面校正板的相機參數校正技術，求得扭曲校正與消除暗角效應所需之相機內部參數，並利用平面投影轉換將影像轉換成鳥瞰視角的影像。接著偵測鳥瞰影像中的角點，藉由角點匹配獲得四張鳥瞰影像間的幾何轉換關係，將鳥瞰影像拼接為一張俯視車輛週遭的全周鳥瞰影像。最後，使用分析各影像的亮度變異，將四張影像亮度調整成一致，再將重疊區域的色彩混合，完成環場鳥瞰影像的接合。本系統的即時運算程序在Intel? Pentium? Core2 Duo 2.66GHz及1.99GB RAM的個人電腦上可達每秒26張。 To park cars in somewhere, drivers usually need to turn their heads or look through the side-view and rear-view mirrors to watch the area behind the vehicles. Due to the limitation of field of view, drivers are mostly unable to see all the area around the vehicles. For the safety of drivers, we present a real-time surrounding bird-view monitoring system for parking assistance. We mount four wide-angle cameras at the front, rear, and both sides of the vehicle to capture consecutive four-in-one images to synthesize the surrounding bird-view images. The studying works consist of off-line and on-line processes. In the off-line processing, we calculate the camera intrinsic and extrinsic parameters and then estimate parameters of distortion model and vignetting model for distortion correction and vignetting compensation. We obtain four bird-view images by computing the homography matrices between every image plane and a virtual image plane which is parallel to the horizontal. Based on the feature points on these images, we register four bird-view images to produce a surrounding bird-view image. Furthermore, we build the lookup table for recording the mapping between the images captured by the wide-angle cameras and the surrounding bird-view image to speed up the on-line processing. In the on-line processing, we use a bi-linear interpolation method to obtain the color value of every pixel of the bird-view surrounding images, and then we compensate for vignetting effect, uniform illumination, and blend the color of the overlapped regions to obtain the seamless images with smooth illumination. Finally, we provide the driver with the surrounding bird-view images to monitor the surrounding area of the vehicle.
