本論文針對車載型攝影機之應用,提出影像穩定化(image stabilization)的技術,主要目的在於補償架設於車上之攝影器材因晃動而造成不穩定的影像序列。在影像穩定化過程中,可以分成三步驟:道路標線的偵測(lane marking detection),攝影機運動的估測(camera motion estimation)及運動的平滑化(motion smoothing)。首先從輸入影像中偵測出兩條道路標線與消逝點(vanishing point)。接著利用兩條道路標線與消逝點來估測攝影機的運動參數,如此不但可以避免一般方法所遇到特徵點擷取與像點對應比對的問題,同時所得到的道路資訊也可用於後續之處理工作。接下來我們再利用運動平滑化來消除不要之攝影機運動以修正影像序列,使其顯得穩定。我們以模擬影像及真實影像來驗証所提出的方法。
The paper presents an image stabilization technique to stabilize the image sequences taken from a camera mounted on a moving vehicle. The proposed technique includes three steps: lane marking detection, camera motion estimation, and motion smoothing. First, we detect two lane markings and the vanishing point from the input image. These two lane markings and the vanishing point are used to calculate the values of the camera motion parameters. The proposed motion estimation can avoid the issues on the feature point extraction and the corresponding matching, and the obtained road information can be applied to the further work. Finally, the motion smoothing is used to separate smooth and non-smooth motion, and then the stabilized image sequences can be obtained. Experimental results for all steps using simulated or real images are presented.