Research And Implementation Of Near Infrared Human Eye Detection And Tracking Based On Android

Iris recognition is the most secure method in biometric identification.Human eye detection of near-infrared images is an important part of iris recognition.However,the traditional near-infrared human eye detection method is very difficult to take both detection speed and accuracy into account in Android.In this thesis,the algorithm of deep learning combined with kernel correlation filter is used to detect and track the human eyes in real time and transplanted in Android.The main work of this thesis is as follows:1.Build a near-infrared face video image database.A camera with near-infrared light is used to acquire face images for detection and face videos for tracking are obtained by using a camera with near-infrared light.The image resolution is 1920x1080,there are a total of 55,000 images including 50,000 training images and 5,000 test iamges.There are a total of 10 videos,each video has a duration of 10 seconds.2.Designed and implemented a human eye detection algorithm on near-infrared images based on convolution neural network.The network is composed of 5 deep separable convolutional layers,13 standard convolutional layers and 2 connection layers.The K-means method is used to optimize the parameters of the anchor in the network,and Softmax and Smooth L1 loss are used for class classification and coordinate regression.The experimental results show that the accuracy of eye detection is 98.8%.3.Studied and implemented a human eye tracking algorithm on near-infrared images based on improved kernel correlation filter.The algorithm uses local binary pattern feature,multi-scale template and output maximum response threshold for tracking optimization.The experimental results show that the precision of eye tracking is 94.1%,the tracking speed is 7 ms/f,and the coincidence rate of human eye boxes is 81.4%.4.Near-infrared human eye detection and tracking based on Android mainly adopts the deep learning framework,NCNN,which is an open source of Tencent.By compiling the source code,the detection model is transformed,and the transplantation of the Android system is carried out in combination with the encapsulated tracking algorithm.The test results show that the final detection speed of the system for human eyes is about 30ms/f,which meets the real-time requirements.