车辆检测算法是汽车辅助驾驶系统中的核心算法之一。为了对摄像机抖动下的目标进行鲁棒跟踪，提出了一种新的基于投影的曲线模型匹配方法，利用投影的完整度、匹配点相对于模型的偏移量期望和方差的加权和作为相似性度量。在此基础上建立模型匹配算法，利用车辆模型匹配进行车辆的检测和跟踪。实验结果表明，该算法具有良好的适应能力，对不同车型、大小、颜色的车辆都能够进行很好的检测和跟踪，并能够在车辆颠簸造成摄像机抖动的情况下保持目标的稳定跟踪。

Robust and reliable vehicle detection is the first step in automotive driver assistance systems. In this paper, a new approach of model matching based on curve projection is presented. The similarity measurement is based on the weighted sum of integrity of projection, offset expectation and variance of matched point. A scheme for vehicle detection and tracking is presented based on model matching. Results for natural traffic scenes demonstrate high reliability of the proposed method.