Human Body Tracking Algorithm Based On Human Face Information In Complex Scenes

With the rapidly development of machine automation,vision principle,which is a core to access surround information,has been attracting researchers.Video object tracking is one of key technology in computer vision,and it is also a hot spot in image processing.Nowadays,it is utilized in human-computer interaction,security surveillance,and imaging guidance.However,challenge scenarios such as illuminations,gestures and postures,and quick moves deter the tracking results.To reach a better performance in the scenarios can make a great step in this field.The existing typical video object tracking algorithm,Tracking-Learning-Detection,has been mentioned due to the excellent performance in long time tracking.The three stage of TLD are tracking,detection,and learning.But the algorithm is apt to lead drift or tracking failures when the tracked object is disguised or disappearance and appearance is up.In this thesis,two improved TLD tracking algorithms based on facial information are proposed to solve the problems caused by object disguises.The main contents are following:1.Robust human tracking via key face information.Facing to drift caused by the TLD,it accumulated classification mistakes.In order to prevent from the drift,this thesis proposes the tacking algorithm using contextual information to integrate more reliable information in this processing.The auxiliary tracking algorithm based on facial key features is proposed,which is able to detect moving human body and faces and update an online trajectory association model.When the failure of human body tacking happens,the algorithm will utilize the auxiliary tracking algorithm.It is proved that the algorithms have a great performance in half or full disguise scenarios.2.Video tracking method based on improved tree model.Based on the facial auxiliary tracking algorithm,Histogram of sparse code is applied for the facial feature extraction stage which improves the accuracy and robustness.The online trajectory association model is update according to the improved algorithm and moving human body.The comparison experiments prove that the proposed algorithm is more robust than the others.