A Hand Gesture Recognition Algorithm For Humanoid Robot And Its Realization

In order to meet the social productive forces' demand for intelligent development,more nations have paid attention to intelligent robot technology.In the meantime,many business organizations get their intelligent robot products into the market.For purpose of upgrading those intelligent robots,the number of sensors and other devices is increasing.Human-computer interactions can't able to meet the demand of intelligent devices.Currently,gesture interaction as an original but innovative interactive mode which is fit to Human-Computer interaction.Gesture recognition as a very important issue of gesture interaction,the implementation of its system is also an important research subject.Based on the research results both abroad and at home,the method of gesture recognition have many defects,mainly includes the following aspects:First,These algorithms are so outdated.There are many traditional machine learning algorithms rather than these most popular Deep learning algorithms;Second,These algorithms needs fixed experimental environment.It is difficult to meet the needs of reality;Third,These algorithms are based on PC platform.They do not to account for embedded platform.To solve the above problems,we carried out the following research work:First,combined with real requirement,this paper sets out design and realization of a method of gesture segmentation based on skin color information and movement information.In consideration of almost all of gesture segmentation method have their own defect,neither of them can fit to various real environment.This paper present a method,which combines skin color information and movement information,is of good robustness against various environment.Then,aiming at the excellent performance of CNNs,it presented an improved CNNs algorithm to solve the difficulty of artificial design features,and redesign the network structure.Experimental results show the performance of recognition and classification are so well.At last,we transplant the gesture recognition algorithm to Nao.