在視訊監控的應用中，影片中車輛的顏色是重要的分析項目之一。由於車輛顏色在影像中的
表現，會受到攝影機所在環境光源的影響，因此在本文中，我們提出一個二階段式的自動化車輛
顏色分析方法。首先，我們以SVM將含有車輛之影像，依其顏色分佈對天候進行分類。第二步，
依照分類後的天候，選擇對應的車色分類器。車色分類器是基於 PLSA(Probabilistic latent semantic analysis，機率式潛在語意分析)方法建構，此方法可自動在影像中，挑選具有車輛顏色代表性的部分進行分析，以分析出正確的車輛顏色。因此，本文提出的演算法，也較不受影像中其他雜訊，例如有部分錯誤的移動物體偵測結果的影響。

The car-color identification is an important issue in the video surveillance. Due to the fact that the car-color is highly dependent on the ambient light condition, a two-step car-color analysis system is
developed in this paper. In the first step, the car image frames are classified by weather condition using SVM. Next, different color classifiers based on the Probabilistic Latent Semantic Analysis (PLSA) model are applied. The proposed algorithm can be used to avoid the false color identification under different weather conditions. Moreover, the algorithm can automatically identify the most significant part of the car-color and classify it correctly. Thus, even with the background noise from moving objects, our algorithm is still robust and
efficient.