- 期刊
混合式行進間移動物體即時偵測技術
Real-time Hybrid Moving Object Detection on Moving Platforms
摘要
本論文提出－基於移動式攝影機之混合型移動物體偵測技術，由於傳統上移動平台所擷取之視訊影像其前景及背景皆同時移動，故無法直接於無處理之影像上建立穩定的背景模型以偵測移動物體。本研究使用「連續畫面差異」技術輔助「背景相減」技術的方式，在背景移動校正補償後的影像上偵測前景物體，即可有效且即時地進行移動攝影機之移動物體偵測。同時利用了「距離轉換(distance transform)」技術，將連續畫面差異偵測結果轉換為距離地圖，對於物體所在的位置，給了一個機率分佈，將此分佈應用至「背景相減」技術中，成為了前景偵測及背景更新之很好的權重值。前景偵測時對於移動物體部分加強權重，而背景更新時將降低移動物體部分的權重，使得移動物體更易被偵測出，此將可改善背景相減技術的缺點，而更穩定地偵測出移動物體。實驗結果顯示，本方法可穩定且正確地於行進間偵測出移動物體之良好外型。
並列摘要
The traditional object detection approaches will have the problem of extracting the foreground objects among the varied background when the camera moves. In order to solve this kind of problem, a new object detection technique is developed in this paper for moving cameras. After performing the motion compensation to adjust the background area among the consecutive frames, the distance map can be derived by the consecutive frame differencing approach. Then, the distance map will be useful in the background subtraction procedure by providing weights for both background area and foreground objects. That is, the moving object will be easily detected in the foreground extraction procedure due to the larger weights, and the background area remain the same condition in the background updating procedure due to the lower weights. The experimental results show our algorithm not only efficiently and correctly detects moving objects, but also precisely extract the objects with better appearance.