Research On Robust Face Tracking Based On Video Sequence

With the rapid development of science and technology and gradual improvement of people's requirements for information security,biometric tracking and recognition technology in intelligent environment has become a research hotspot in the field of pattern recognition and human-computer interaction.Human face is a special class,because it plays an important role in understanding human activities in many usage scenarios.As an important research direction of computer vision,face tracking has always been a research hotspot in computer vision.Face tracking is often disturbed by various factors such as occlusion,deformation and scale transformation,which often leads to tracking drift and ultimately leads to tracking accuracy reduction or even tracking failure,failing to meet the expected requirements.On the basis of compressive tracking,this paper studies face target tracking from three aspects including feature fusion,adaptive feature weighting and improved robust tracking method.The main work is as follows:1.The problem of multi-feature fusion representing human face information is studied.In view of the problem that only a single haar-like feature is used in the compressive tracking method,which cannot accurately describe the facial feature information,a face compressive tracking method based on extended haar-like feature and LBP feature is proposed.Two kinds of features are used to enrich the facial feature information,and the sampling strategy from coarse to fine is adopted.Firstly,extended haar-like feature is used to carry out coarse tracking of the face target position to obtain the approximate position of the target.Then the accurate sampling is carried out near the target location area,and the LBP feature is used for fine tracking,and finally the optimal location of the face target is obtained.2.The problem of feature weighting and adaptive template updating under compressive tracking framework is studied.The classifier of compressive tracking method adds many different sample features directly,and does not consider the impact of different sample features on the target classification effect.In this paper,the idea of weighting classification is adopted.Different weights are assigned to different features based on their different contributions to the classification effect,and larger weights are assigned to the reliability features,so as to improve the classification accuracy of classifiers.At the same time,aiming at the problem that fixed learning rate is used in the compressive tracking method to update the template,which cannot adapt to the target transformation and cause tracking drift,the updating rate of the template is adjusted adaptively through the target interference judgment,so as to improve the tracking accuracy of the target.3.The robust face compressive tracking method based on kalman predictor is studied.Aiming at the problem that the compressive tracking algorithm has weak anti-noise ability and is easily disturbed by external factors,which leads to the decrease of tracking accuracy,a compressive tracking method based on kalman predictor is proposed.After the face target is judged to be interfered,kalman filter is introduced to estimate the correct position of the target,with this location as the target,the subsequent face target tracking is continued.