为了解决固定摄像机下车辆跟踪过程中阴影对检测的影响，提出一种改进型码本模型的车辆检测方法。该方法直接对YUV空间的车辆序列进行处理，将采样到的背景值聚类成码本，对于新输入的像素值与其对应位置的码本作比较判断，提取出前景区域。车辆跟踪中采用Kalman预测的方法来处理车辆遮挡问题。实验结果表明，本算法可以从复杂交通场景图像序列中快速有效地检测出运动目标，能较好地处理阴影、高亮、遮挡和背景变化等问题，且计算复杂度小，能满足实时跟踪的需要。

In order to overcome the effect of shadow in process of vehicles tracking under stationary camera, we present an improved codebook model detection algorithm. This method deal with vehicles sequences directly in the YUV Color Space, and the sampled background values are quantized into codebooks. Input pixel values of new frame are compared with the codebooks to identifying foreground areas. The Kalman Prediction method is used for vehicles tracking which can deal with occlusion. Experiments show that this algorithm can detect moving objects in complex traffic scenes effectively and rapidly. The proposed method can handle shadows, highlights, occlusion and the change of background all of which make this method efficient in both computation and the needs of real-time tracking.