基于视频的车辆检测在智能交通系统中有着重要的实用价值.提出了一种在复杂背景中检测运动车辆的方法，针对传统光流法在阴影、边界和遮挡的地方灰度守恒和光流场平滑性假设不再成立这一问题，引入前向-后向光流方程，计算其Hessian矩阵，并把Hessian矩阵的条件数与Lucas-Kanade光流法中的加权阵相结合，有效地消除了局部邻域中不可靠的约束点，同时进一步提高了光流约束方程解的稳定性.实验结果表明：该方法检测情况稳定，检测准确率高，检测效果好.检测结果可作为智能交通系统中高层交通管理和车辆控制的基础.

As the video vehicle detecting　has a great importance to the intelligent transportation system, a new moving vehicle detection method based on optical flow in complicated background was presented. To deal with the hypothesis that the smoothness of optical flow and the conservation of gray intensity don’t come into existence on the condition of shadow, boundary and overlap, a new forward-backward optical flow equation is introduced and the Hessian matrix of equation is computed. Then the conditional number of the matrix is combined with the associated matrix of Lucas-Kanade’s optical flow method to eliminate the restriction points which are not credible within local neighborhood. Experimental results show that the proposed method helps to obtain satisfying detecting results due to its good stability and high detecting accuracy. The results can be used as the basis of advanced vehicle control and traffic management in intelligent transportation systems.