The Research And Implementation Of Human Face Detection And Tracking Based On Complex Background

Cause the human face provides a wealth of valuable information,so in areas such as the intelligent human-computer interaction,computer vision research and Intelligent monitoring,face detection and tracking technology has become a hot research topic.Face detection is the process for determining the position and size of the human face from the image.Face tracking is the process to determine the moving trajectory and position changes of the human face from an image sequence. This paper studies the human face detection and tracking algorithm in complex scenes, and also describes the algorithm improvements.In regard to face detection problems in the complex background,this paper uses Adaboost face detection algorithm by Viola for the foundation, explains the process of the face detection algorithm,describes the shortages of the algorithm,and proposed some improvement strategies for these deficiencies.1) For the traditional Adaboost face detection algorithm, Adaboost classifiers can not construct a minimum error rate classification. And the traditional Adaboost algorithm treats sample classification as tow symmetrical classification problems.However, towards the cascade detector,we should ensure that each layer classifier has low false detection rate for positive samples,and then we can ensure that the final layer classifier has low false detection rate for positive samples.Therefore, this paper will use the AD Adaboost face detection algorithm for the process of classifier training.This algorithm ensures a high detection rate by the classification,and the false detection rate significantly decreased,thereby this algorithm improves the face detection performance.2) Towards the traditional Adaboost face detection algorithm,the cascade detector is one-dimensional cascade detector,this cascade structure can not effectively solve the problem that the vectors of face feature set distributes multimodal.This paper proposes the application of the K-means algorithm,divides the face sample cluster into K subsets,trains the strong classifier for each subset,and construct the new two-dimensional cascade detector,thereby improves the performance of the cascade detector. The experimental results show that under the same conditions, the algorithm can improve the detection accuracy and the detection speed.For the face tracking problems in the complex background,this papaer describes the basic principles of CamShift algorithm and Kalman filter algorithm,and then this paper describes the application of this tow algorithm in face tracking field.This paper did the human face tracking experiment by CamShift algorithm,and found the algorithm is easy to loss target when tracking in the situation which the face target moves rapid or face blcoked.This paper use the algorithm combine CamShift algorithm with Kalman filter algorithm for face tracking.The experiment results show that the improved algorithm can improve the tracking performance under the scene which the face target moves rapid or face blcoked.