Human Motion Recognition Based On Background Cut Off And HMM

As the develop of science and technology, artificial intelligence for reconnition has largely developed. Currently,people pay more and more attention on the research of human motion recognition. Human action recognition has a wide range of applications in intelligent surveilance,human-computer interaction,virtual reality,motion analysis and other fields has a wide range of applications.Based on the research of human motion recognition home and abroad,this paper improved the traditional HMM recognition algorithm,.And this paper presents a new method based on the background cut off and EHMM combination of human action recognition method.The main contents are as follows:Firstly, the action sequences for video adopted gaussian mixture model method for target detection. Secondly,cut off background for each frame of video images,extract the foreground silhouette of human action,to find out body silhouette,preraring for the next action.Thirdly, for the feature extraction the body divided into five parts,containing head, arms, waist, thighs, legs and feet.Each part as a super state, every super state containing several parts as embedded states. Fourthly, carries on the feature extraction, form the observation vector sequence. Fifthly, training EHMM model, and deposite feature vectors into the EHMM feature library.Last of all, to compare featurre vectors in the EHMM feature library and the featurre vectors of the need recognizing action, then find out the maximum probability matching feature vector, and thus recognize the human action.In the pilot phase, the two sets of different experiments were designed.In the first set of experiments, the article selected 10 movements of.common life. At the second set of experiments, comparing the method with CRF, HMM to prove the effectiveness of this method. The final experiment results show that this method in dealing with video data have a relatively good effect.