由行進中攝影機取得之光流場模型來進行移動物體偵測

Abstract

移動物體偵測是電腦視覺系統裡重要的一環，特別是在運輸系統中。移動 物體偵測已經被廣泛研究，文獻上有許多方法被提出來。可是，當視訊是由移動 中的攝影機所擷取時，移動物體偵測是很有挑戰的課題。因為在這種情況之下， 不只是真實移動物體，靜態物體也會呈現相對運動狀態。因此，這種情境下的真 實運動物體偵測是很困難的工作。 為處理這個困難，我們嘗試使用光流場的方法來偵測真實移動物體。第一 步，我們在一輛車輛四周安裝魚眼攝影機，以擷取移動物體與靜態環境的視訊。 第二步，我們是用光流場估測來計算視訊中的移動模式。最後一步則是在光流場 中進行移動物體與靜態環境的分割。因此，我們的方法可以在視訊中擷取真實移 動物體。 在這項工作中，我們的方法被應用在兩種情況中來進行移動物體偵測，包 括簡單情境與複雜情境。根據我們的實驗結果，我們的方法在簡單情境中有很好 的移動物體追蹤與偵測效能，而在複雜情境中則會發生一些偵測錯誤的情況。

Moving object detection is an important part of computer vision systems, especially in transportation systems. Moving object detection has been widely investigated and many algorithms have been proposed in the literature. However, moving object detection becomes more challenging when the image sequence is captured by a moving camera. In this case, not only real moving objects, but also static objects can be regarded as moving objects. Therefore, it is difficult to identify real moving objects in this situation. To tackle this problem, we attempted to detect the real moving object by using an optical flow method. In the first step, we used the fisheye camera placed around a moving vehicle for recording image sequences of static and moving objects in the environment. In the second step, the optical flow estimation is used to detect patterns of movement in the image sequences. The last step is to segment the moving and static objects. Therefore, our method can extract the real moving object in the image sequence. In this work, our method has been applied to two cases, including simple scene and complex scene, for moving object detection. According to our experiment results, our method showed a good performance for tracking and detecting moving object in the simple scene. On the other hand, some false detection of the moving objects happened in complex scene.