The Research Of 3D Space Handwritten Character Recognition Method Based On Leap Motion

Handwritten character recognition technology has a very important role and broad development prospects in pattern recognition,artificial intelligence,information security and human-comp ut er interaction.So far,handwritten character recognition technology is mainly used in automatic sorting of mail,entry of statistical reports,bank note processing,account password recognition,humancomputer interaction and so on.Many fields require relatively high accuracy of handwritten character recognition.As an important step in 3D space handwritten character recognition,dimensiona lit y reduction is a key step to simplify the recognition process and reduce the cost of calculat io n.Therefore,its research has become especially important.First of al,this paper analyzes and researches handwritten character recognition and related contents with handwritten character recognition.And common dimensionality reduction algorit hms(PCA,LDA,LLE and LPP)are introduced.According to al-round comparison,the existing shortcomings and deficiencies are summarized.We find that the direction of 2D characters proposed by existing dimensionality reduction method which maintains the global structure of data will flip randomly.For the sake of reducing the error rates of 2D images,the direction of 2D images need to be adjusted before images recognition.Then,in order to improve the shortcomings and deficiencies of the above algorithms,this paper presents a new algorithm: dimensionality reduction algorithm based on the longest trajectory projections.The algorithm can not only simplify the recognitio n process,but also improve the recognition rate of handwritten characters.The algorithm steps are as follows.Firstly,getting the 3D coordinates of moving fingertip and connecting the 3D fingertip coordinates in turn to generate the 3D trajectory.Next,projecting all points on 3D trajectory to XOY,XOZ,YOZ planes and forming the 2D trajectories.Finally,calculating the length sum of the 2D trajectories on the three planes.And selecting the plane that have the maximum 2D projection trajectory length sum as the optimal projection plane.Through the experiment and analysis of the algorithm,it can be found that the proposed algorithm has simple principle and concise steps.The 2D images obtained by the proposed dimensionality reduction algorithm not only preserve the global structure information of the origina l data,but also have a fixed direction and a good visual effect.In addition,this paper uses the proposed dimensionality reduction algorithm and PCA-based dimensionality reduction algorithm to reduce the dimension of 3D space handwritten characters.The experimental results show that the dimensiona lit y reduction algorithm has better visual effect and stable result,and the character image will not be flipped randomly.However,the orientation of the 2D character images processed by the PCA-based dimensionality reduction algorithm will be randomly flipped,which will seriously affect the subsequent feature extraction and the final recognition effect.The proposed dimensionality reduction algorithm is more suitable for dimensionality reduction than PCA-based dimensionality reduction algorithm for 3D space handwritten character recognition.Finally,in order to further verify the feasibility and effectiveness of the proposed algorithm,the character images obtained by the proposed dimensionality reduction algorithm and PCA-based dimensionality reduction algorithm are respectively recognized and the recognition results are compared.The recognition results show that the average recognit ion rate of the character images obtained by the proposed dimensionality reduction algorithm is larger than the average recognit io n rate of the character images obtained by the PCA-based dimensionality reduction algorithm.The proposed dimensionality reduction method in this paper can obtain 2D images with fixed direction,and the recognition rate of 3D space handwritten characters can reach to 96.5% without the direction adjustment algorithm.