Research And Design Of Face Recognition And Tracking System Based On Adaboost And SIFT Algorithm

Face recognition and tracking system is to detect faces and track the target face from imagesequences containing complex background by using image processing technology. Face recognitionand tracking technology was widely used in social production and people’s livelihood, such as videosurveillance, human-computer interaction, expression analysis and so on. In recent years, with thedevelopment of image processing technologies, research on face recognition and tracking attractsmore and more researchers’ attention.With such considerations, a human-face detection and tracking system based on Adaboostalgorithm and SIFT algorithm is proposed in this paper. The whole system can be divided into fourparts: image preprocessing, face detection, face recognition, and face tracking. At the first step, asuitable preprocessing to the input image will be performed. Secondly, the face detection algorithm isused to detect faces in the input image. Then face recognition which matches the target face in thedetected faces is performed to find out the location of the target face. At last, the system is to track thetarget face in the image sequence.In the face detection part, an improved Adaboost machine learning algorithm which is based on theHaar_Like characteristics of the image is proposed. On the basis of the original algorithm, animprovement for the sample weight update rule is made in this paper. The experimental results haveshown that the robustness and effectiveness of face detection is improved in this way. In the facerecognition step, a face recognition algorithm based on SIFT is used. The SIFT algorithm is invariantto image orientation, image scale and partly to affine distortion, so it can solve the problems such asrotation, image size inconformity that appears in human face recognition. In the face tracking part, anapproach that does the face detection and face recognition mentioned above in every frame of theimage sequence is used.With the system discussed above, plentiful experiments have been done to explore the performanceimprovements, the selection of the preprocessing algorithms and system parameters according to thesystem performance with the help of OpenCV, VC++and Matlab. Through the analysis of experimentresults, we have demonstrated that improved performance have been achieved with the improved facedetection algorithm, proposed reasonable preprocessing precept and appropriate system parameters.