Research Of Face Detection And Tracking Algorithm Based On Kernel Correlation Filtering

Face detection under natural conditions is one of the most basic and core research topics in the field of computer vision.This subject not only has important theoretical research significance,but also has engineering application value in aspects such as assisted driving system,system monitoring,and terminal brushing.After related research in recent years,although there have been many effective face detection algorithms,it is still quite challenging to design a face detection algorithm with high real-time and accuracy.During the tracking process,the human face will be affected by the external environment such as light and occlusion,as well as the effects of external rotation such as out-of-plane rotation,in-plane rotation,expression change,rapid movement,and scale change.The above changes in appearance will cause the task to fail when the face detection tracking model is unable to cope with it.Therefore,how to design a real-time and robust face detection tracking algorithm is still a hot topic in the field of face detection tracking.This article works from the following aspects:1)By analyzing the advantages and disadvantages of the Dlib face detection algorithm and the kernel correlation filter tracking algorithm,this paper proposes a face detection and tracking framework based on kernel correlation filtering.Specifically,this method fuses the Dlib face detection algorithm and the kernel correlation filter tracking algorithm by the face verification method to improve the real-time performance of the face detection tracking algorithm.Aiming at the problem of face scale adaptive updating and tracking model robust updating,the SDM-based face detection tracking algorithm is proposed to integrate the SDM face alignment algorithm into the face detection tracking algorithm framework.On the one hand,SDM can use the face feature points after alignment to achieve the face of the scale adaptive;the other hand,the results of the tracking of SDM face alignment,the use of accurate face position and scale information to update the tracking model to ensure the robustness of the tracking model.Finally,experimental verification is performed on the collected face video dataset.The face detection tracking algorithm based on SDM surpasses the mainstream face detection tracking algorithm in terms of real-time performance and accuracy.2)Vision-based face detection tracking technology is a type of long-term tracking technology.In order to ensure that faces fail to be detected in time after a failure in tracking,this paper proposes a face verification method based on Color-naming features.In this method,a Color-naming-based histogram feature is designed and used as a face verification model.The face verification method includes a learning process and a verification process.After the initial stage or when the object has a long appearance change,the verification model needs to be studied.After the learning is completed,the tracking result is analyzed and verified using the learned face verification model.This method can antomatically adjust the face verification model to cope with changes in the appearance of the face.Further,we propose a face detection tracking algorithm based on adaptive verification.This algorithm incorporates an adaptive face verification function to ensure the robustness of the algorithm.Finally,experimental verification is performed on the collected face video dataset.The face detection tracking algorithm based on adaptive verification surpasses the mainstream face detection tracking algorithm in real-time performance and accuracy.