Research And Implementation Of Eye-Computer Interaction Model Based On Improved YOLO-v3 Algorithm

With the continuous development of Artificial Intelligence in recent years,the way of Human-computer interaction have emerged one after another.Among them,the eye-computer interaction is a human-computer interaction method based on eye movement.Its principle is mainly to realize the positioning of the human eye by detecting the position of the human eye,and use the extraction of the human eye feature to distinguish the eye behavior,and realize the control through the eye behavior.This kind of interaction can free hands in a specific environment,and has very important research and application value for the physically disabled,severely paralyzed,vehicle driving,and in the military field.As the embedded application environment becames more and more extensive,smart and portable wearable eyeball interactive devices have become one of the current research focuses.With the continuous development of deep learning,Convolutinal Neural Networks(CNN)provide a strong role in the extraction of image features,and provide key technical support for eye detection in eye recognition technology.Therefore,based on the eye machine interaction technology,human eye recognition technology and target detection theory of YOLO-v3,this paper proposes an eye-computer interaction model and real-time eye behavior recognition algorithm based on the improved YOLO-v3 algorithm,and the eye control command Output.First of all,This article explains different types of target detection algorithms based on convolutional neural networks,and understands and analyzes the working principles and applications of various algorithms.The advantages and disadvantages of various target detection algorithms are analyzed and compared,and the YOLO-v3 algorithm is mainly studied.Secondly,an improved YOLO-v3 eye interaction model is proposed.Firstly,researchers use the infrared camera to capture face images,and the data set is collected.Then,according to the YOLO-v3 model,delete the feature detection layer and increase the number of layers of the shallow network,and the initial priori box is selected by K-means clustering algorithm,which improves the fine granularity of network pixel feature extraction and speeds up the detection speed.Finally,the eye-machine interaction model was constructed by combining the extraction method of human eye characteristic parameters,eye behavior recognition algorithm,and eye behavior control command output method.The results shows that the improved YOLO-v3 algorithm can detect human eyes with an accuracy rate of 99.9% and the recognition speed of22.8FPS.Compared with the original YOLO-v3 algorithm,m AP was improved by about 7.01%,and the detection time was shortened by about 11 percentage points.Finally,real-time eye behavior recognition is realized by using eye feature parameter extraction method and eye behavior recognition algorithm.Under the same conditions,the eye with glasses and the eye without glasses are identified to verify the effectiveness and accuracy of the eye machine interaction model.The experimental results show that the recognition rate of different eye behaviors is 91.30%.At the same time,the simulation test of eye control signal output is carried out,and the eye control command is found correctly.