基于隐马尔可夫模型的运动目标轨迹识别  

Trajectory recognition of moving objects based on hidden Markov model

作　　者：潘奇明[1] 程咏梅[2] 

机构地区：[1]电子科技大学中山学院自动化系,广东中山528402 [2]西北工业大学自动化学院,西安710072

出　　处：《计算机应用研究》2008年第7期1988-1991,共4页Application Research of Computers

基　　金：国家自然科学基金资助项目(60372085)

摘　　要：引入改进的隐马尔可夫模型算法,针对真实场景中运动目标轨迹的复杂程度对各个轨迹模式类建立相应的隐马尔可夫模型,利用训练样本训练模型得到可靠的模型参数;计算测试样本对于各个模型的最大似然概率,选取最大概率值对应的轨迹模式类作为轨迹识别的结果,对两种场景中聚类后的轨迹进行训练与识别。实验结果表明,平均识别率分别达到87.76%和94.19%。Using modified hidden Markov model, firstly, aiming at the complex degree of the objects＇ trajectories in real scene, the models were built for every trajectory pattern, and the training samples were used to get the credible parameters of the model. Finally, the maximum likelihood probability of test samples were computed to all of the trained model, the maximum value was saved and the corresponding model was the recognition result. Then train and recognize the samples clustered, and average recognition rate reach 87. 76 % and 94. 19% respectively,

关 键 词：轨迹识别 运动分析 行为模式 隐马尔可夫模型 

分 类 号：TP242.62[自动化与计算机技术—检测技术与自动化装置]