本项目针对多个运动目标相互遮挡情况下的跟踪问题，提出采用遮挡分层模型为遮挡目标建模，建立起关于目标位置、尺度及目标间遮挡关系等状态的概率模型，从而将多目标跟踪问题转化为该概率模型下的目标状态最优估计问题。为充分利用不同特征之间的互补性，以更好地适应跟踪过程中场景的变化，提出利用目标外观与光流特征的动态组合来描述目标，构造反映目标状态的似然概率。针对目标状态空间维数随跟踪过程中目标数量的变化而变化，且状态空间中参数类型多样的特点，提出利用包含跳跃和扩散两种状态转移方式的马尔可夫链蒙特卡罗（MCMC）方法进行目标状态最大后验概率寻优，以获得全局最优的跟踪结果。为了加快优化函数的收敛速度，提出利用目标历史状态信息和领域知识来构造MCMC状态转移函数。在此基础上，与目标状态预测技术相结合，以实现各种遮挡条件下的多目标跟踪，为地空/空空导弹精确制导、智能视频监控等应用提供技术支撑。

As to the multiple objects tracking under occlusion, we proposed the occlusion layer model to describe the occlusion relation among the objects and build a probability model about the object position,scale and object occlusion relation so as to transform the multiple object tracking problem into an optimal estimation problem about this probability model. In order to make full use of the complementarity of different image features to adapt to the varying scene, we proposed to utilize the object appearance and optical flow to represent the objects jointly, moreover, the weights of the different features are adjusted dynamically. For the dimension changing with the number of object and various object status types, we proposed to optimize the maxmimum a posteriori estimatioin(MAP) using Markov Chain Mento Carlo(MCMC) sampling technique including the transformtion mode of jump and diffuse to obtain the global optimal solution. In oder to optimize the velocity of the convergence, the historical information about the scale and domain knowledge are employed to construct the status transfer functions. Base on the above fact, we introduced the predicting technique so that our tracking system can track the occluded objects under different situation and further be applied into the video surveillance, the precision guidance of missile and so on.