Research On Algorithm Of Multi-view Human Action Recognition Based On Image Set

Human action recognition from videos is a hot research topic in computer vision,which has been applied into surveillance system and human machine interface.Many researchers had paid attention into this field,and obtained satisfying performance.However,many traditional action recognition algorithms are based on single view,which will be affected by light,shade,high variability of appearance and shapes.Thus,many researchers put their attentions on multi-view action recognition,but how to mine the relationships among different views? Since video face recognition algorithms based on image set have proved that image set algorithms can effectively mine the complementary properties of different views image,and achieve satisfying performance.Thus,multi-view action recognition algorithms based on image set is proposed.This contribution of this paper is as follows:1)Multi-dimensional human action recognition model based on image set and group sparsity is proposed.Firstly,we extract dense trajectory feature for each video;Secondly,we construct the shared codebook by k-means,and then Bag-of-Word(BoW)weight scheme is employed to code dense trajectory feature by the shared codebook;Lastly,multi-dimensional human action recognition model based on image set and group sparsity is trained.Large scale experimental results on three public multi-view action3 D datasets,show that multi-dimensional data is very helpful,and the proposed scheme based on image set can further improve the performance.What is more,when group sparsity is added,its performance is comparable to the state-of-the-art methods.2)The influence of the number of samples in image set is evaluated.Comparing face dataset and human action dataset,we can observe that the number of each image set in video face recognition is requested to have several ten to several hundred samples.But,in multi-view action recognition,we only have 3-5 samples in each image set.Thus,we will evaluation the effect of number of samples in image set.Firstly,we will perform large scale experiments on public face datasets by the state-of-art algorithms.With the change of the number of samples in image set,the experimental results show that the argument of the number of gallery set or query set is very helpful for classification.And then we do further verification tests on human action datasets,where we also can observe the same conclusions.3)Reverse testing image set model(called RTISM)based multi-view human action recognition is proposed.We firstly extract dense BoW feature for each video as same as multi-dimensional human recognition based on image set and group sparsity;Secondly,for each query set,we will compute the compound distance with each image subset in gallery set,after that,the scheme of the nearest image subset(called RTIS)is chosen to add into the query set;Finally,RTISM is optimized where the query set and RTIS are whole reconstructed by the gallery set,thus,the relationship of different actions among gallery set and the complementary property of different samples among query set are meanwhile excavated.Large scale experimental results on two public multi-view action3 D datasets,show that the reconstruction of query set over gallery set is very effectively,and RTIS added into query set is very helpful for classification.What is more,the performance of RTISM is comparable to the state-of-the-art methods.