透過您的圖書館登入
IP:18.223.106.100
  • 期刊

以3D物件模型為基礎之車輛偵測演算法

A 3D Object-Model-based Car Detection Algorithm

摘要


本論文提出一種影像式車輛偵測演算法,此方法利用-立體3D車輛模型,套用至欲偵測的物體區塊影像上並旋轉重組物體影像區塊。然後對此重組影像進行水平邊緣特徵偵測,再將水平邊緣梯度影像投影至垂直方向,得出物體水平邊緣之分佈情況作為特徵值,最後使用支持向量機(SVM)以大量樣本進行訓練及辨識,並應用至固定停車格上的停車狀態偵測以及確認所偵測之移動物體影像是否為車輛。實驗結果顯示,在固定停車格上的車輛辨識率達到98%,移動車輛偵測率約為88%。

並列摘要


This paper proposes an image-based 3D object-model-based car detection algorithm. We first use a 3D car template to fit and rotate the detected image blob and recombine the image blob. We then detect the horizontal edge features of that image blob and project them to the vertical line to obtain the histogram of the edge features. Lastly, we use the support vector machine to train the car model and use it to recognize the objects in the fixed parking spaces and the moving objects in the parking lot. Experimental results show 98% recognition rate in the parking space and 88% for the verification of the moving objects.

延伸閱讀