Research And Implementation Of Eye-Computer Interaction Technology Based On Improved YOLOv5 Algorithm

In recent years,artificial intelligence has continued to be popular and developed rapidly,and human-computer interaction technologies in smart devices have emerged one after another.As a more efficient and natural human-computer interaction technology,eye-computer interaction has attracted widespread attention.With the development of convolutional neural network,the target detection algorithm based on deep learning has become the mainstream solution for target detection tasks in various fields.It has the characteristics of high detection accuracy,good real-time performance and strong robustness,and can provide key technical support for eye state detection in eye-computer interaction technology.In the actual application of eye-computer interaction technology,due to the influence of factors such as individual differences in user eye shape,individual differences in user eye behavior habits,and the complexity of the application environment,the difficulty of eye state detection is greatly increased.Therefore,this paper establishes an eye-computer interaction technology based on the improved YOLOv5 algorithm for the above problems.The main work is as follows:(1)This paper proposes an eye state detection model based on improved YOLOv5.First,the external eye camera of smart glasses captures human eye images,and create a human eye state data set based on the three eye state labels "open","closed" and "squinting";secondly,based on the YOLOv5 algorithm,the backbone network is added Coord Attention attention mechanism,improved Mosaic data enhancement method,and improved SPP module,so that the algorithm can obtain accurate position information and region of interest in the spatial direction while obtaining feature information between channels,and speed up the training speed,improve the robustness of the training model;finally,the human eye state data set based on the improved YOLOv5 is used to train the eye state detection algorithm,and the eye state detection model is obtained.Experiments show that the detection accuracy rate of the trained eye state detection model is 99.6%,and the detection speed is 628.93 FPS.Compared with the original YOLOv5 algorithm model,the m AP of the eye state detection model is increased by about4.29%,and the detection speed is increased by about 9.43%.(2)This paper develops eye-computer interaction technology based on improved YOLOv5 algorithm.Firstly,an eye behavior discrimination algorithm is designed according to human eye behavior habits.Secondly,an eye state detection model based on the improved YOLOv5 algorithm is used to detect the user’s static eye state information "open","closed","squinting".Finally,the eye behavior discrimination algorithm combines the static eye state information,discriminate the user’s dynamic eye behavior "gazing","closing","squinting","double blinking" and complete the development of eye-computer interaction technology.(3)Through occlusion and stretch tests,shading and contrast tests,it is proved that the eye state detection model based on the improved YOLOv5 algorithm is more robust than the model based on the original YOLOv5 algorithm.Using the RKNN Toolkit development tool and the artificial intelligence computing stick RK1080,the eye-computer interaction technology based on the improved YOLOv5 algorithm was transplanted to the Raspberry Pi 4B embedded platform for experiments.The average recognition accuracy of the four eye behaviors reached96.55%,and the detection speed reached 78.66 FPS,this technology has high detection accuracy,good real-time performance,strong robustness,and good human-computer interaction performance.