一种基于改进PLSA和案例推理的行为识别算法  

Novel Action Recognition via Improved PLSA and CBR

作　　者：涂宏斌[1] 岳艳艳[2] 周新建[3] 罗锟[1] 

机构地区：[1]华东交通大学轨道交通学院,南昌330013 [2]华东交通大学国际学院,南昌330013 [3]华东交通大学载运工具与装备教育部重点实验室,南昌330013

出　　处：《计算机科学》2017年第6期283-289,共7页Computer Science

基　　金：华东交通大学基金(2003414090);国家自然科学基金(51268015);江西省自然科学(20122BAB206027)资助

摘　　要：针对行为人发生的行为因遮挡或者自遮挡可能导致行为歧义性的问题,提出基于改进PLSA和案例推理算法的行为识别方法。该算法既可以克服传统PLSA算法中生成式模型对观察特征序列的独立性假设会导致过拟合的缺点,又可以消除由于遮挡等原因引起的歧义性带来的识别精度降低问题。实验表明该方法能有效地提高人体行为识别准确率。In order to recognize the ambiguous action meaning in the same scene for occlusion,the improved PLSA and CBR algorithm are used to recognize the simple action according to the space-time interest point.The improved algorithm can not only overcome the shortcomings of the over fitting in the traditional PLSA algorithm owe to the generative model independence assumption of observation sequences,but also decrease recognition ambiguity.Experiments based on the proposed method are executed on the public databases such as KTH,Weizmann,UCF sports and the selfbuilding database.The results show that the proposed method has the performance of validity and effectiveness.

关 键 词：行为识别 歧义性行为 时空兴趣点 PLSA 案例推理 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]