Human Gait Recognition Based On Decision Fusion For Dynamic Video

With the rapid development of Chinese economy,people have a higher demand on social security and stability.Especially on the public security and social security,requirement of identification accuracy has a higher demand.Ordinary identity recognition and eye recognition have many problems and defects,therefore,identity recognition based on computer vision emerges because of requirement.As one of identity recognition,Gait recognition is a more advanced method.Gait recognition is not easy to hide,not easy to camouflage characteristics,and in the remote,it is a common method of identification.Therefore it was given more attention and favor.Gait recognition includes three parts:pre-processing,feature extraction and classification.This paper is about a human gait recognition based on dynamic video and the main research work is divided into the following several aspects:Having a pre-treatment of frames for the continuous gait video and via background subtraction method to extract dynamic foreground.In the feature extraction,it discussed two methods in this paper.One is HU invariant moments and the other is Radon transformation.Having HU invariant moments and having Radon transformation for each frame at the same time.The two groups of extracted values serve as the data of gait and at the same time.The GEI image of HU moment invariant are extracted.Then these data was fused with the Radon transform data as the third group of characteristic data.In the process of decision fusion recognition,each kind of recognition result is used as one vote and the results of the votes are counted as the final result.In order to verify the feasibility of this method and the accurate of recognition.the Analyze the results and compare the HU invariant moments and Radon transformation with the fusion identification results.At the same time,the final decision recognition accuracy was recorded.Compare these four methods and verify the accuracy of this method,meanwhile,compare the accuracy of this method with other traditional algorithms.In order to unify the the experimental data processing,CASIA database which was founded by the Chinese Academy of Sciences Institute of automation was used.Perform the experiment and evaluate these data.Through the comparison of experimental procedures and results,this method which was verified can improve the recognition accuracy and has advantage.The desired recognition effect can be achieved by this method.