Number-sign Hand Posture Recognition Based On Arm-area Removing

With the development of computer science and network technology.the communication between people and computer has become more and more frequent. In order to meet the need of naturality of communication,new technologies are needed in the field of Human-Computer Interaction(HCI).Compared with conventional HCI ways such as keyboard-based and mouse-based HCI ways,the HCI way based on gesture recognition is more natural and simple.The application of gesture recognition is wide, such as remote control by gesture and sign language recognition,and gesture recognition is an important research subject of Human-Computer Interaction.In our daily life,the most commonly used gesture is number-sign hand gesture.and the static one is used more frequently and widely than dynamic one.Considering the differences in hand postures among countries,ten static hand gestures are defined(ranging from hand gesture zero to nine) in this paper.However,differences in arm and arm length will cause information redundancy and characteristical diversities respectively.In this case,this paper just focuses on how to remove the arm.In order to assess the validity of the arm removing method,three situations are considered--no bare arms,a small bare arm and bare arm more.On this basis,a number-sign hand posture recognition algorithm based on arm-area removing is introduced in this paper.1.Existing arm removal methods are mainly based on border width, and search for the minimal value of border width to locate the wrist line,which requires the arm vertical or the wrist line perpendicular to the principal axis of hand area.If the arm is rotated or the wrist line is parallel to the principal axis of hand area,the performances of these methods are not very good.In oder to solve the problem,this paper proposed a distance transformation and inner and outer layer cutting line-segment algorithm of arm-area removing. 2.This paper introduces a method of calculating the number of fingertips based on arm-area removed and the circle of palm center.3.After removing the arm.method that uses Hu moments and fingertips number as features and the BP neural network as a classifier for hand posture recognition is proposed.Through the results of removing the arm area experiment,computing the number of fingertips experiment,and hand posture recognition expriment on the self-built number-sign hand posture picture library,the proposed arm-area removing method has good robustness,which can adapt to the three arm situations,and changes in angle and size well.In addition,method of calculating the number of fingertips based on arm-area removed has high accuracy.At last,method using both Hu moments and number of fingertips as features outperforms the method using Hu moments as features only in recognition rate.