Research On Face And Gesture Detection And Face Tracking

Face analysis is a typicial problem in the field of computer vision.It includes face detecion, face recognition and expression recognition ect. Generally speaking, face and gesture detection, face tracking, which are involved in pattern recognition, intellegent human computer interaction, computer vision and cognitive sciences, form the basis of face analysis. And they have have wide development and application foreground in the field of public safety, video monitoring, and CBIR(content-based image retrieval, especially in the field of human computer interaction.The thesis mainly aims to research on face and gesture detection and face tracking. Firstly the thesis summarizes the main methods and techonologies about face and gesture detection and face tracking. Then the face detection method based on Adaboost, the gesture detection method based on face feature triangle and the face tracking detection based on Camshift are introduced in detail. Finally the intelligent video monitoring system is designed, which can finish the task of auto face detection and tracinging. The major contributions of this thesis are summarized as follows:The face detection method based on Adaboost uses rectangle feature to train cascade boosted classifier. Because the cascade boosted classifier's detection window scans the input image several times in diferent size, it may mark the face region more than one time. A target sequence self-check algorithm is proposed. It deletes the unwanted face mark according to whether the rectangles are covered. Rectangle feature is not the basis feature of human face. In the natural complex environment, false positive rate rises. A face verification algorithm based on Gaussian pyramid and D-S theory is proposed. The algorithm detects every layer of Gaussian pyramid, and integrates all the results to get the correct face regions.The face detection method based on Adaboost can be used in many target detection application. The thesis uses the method to train face feature classifier, to get the face feature points and build face feature triangle that uses eyes and mouth as vertexes. According to the face geometrical property, we can estimate the parameters of face gesture.Meanshift is used to track face. And then an algorithm combined Adaboost and Camshift is proposed to finish the task of auto face detection and tracking.It can be used to design an intelligent video monitoring system.Camshift is an improvement on Meanshift, and it is a rubust statistical method which is realized by searching maximum statistical distribution.Intelligent video monitoring system captures the video by PTZ camera, detects the face frame by frame automatically, until get the face region. The face region is delivered to Camshift, which converts image from RGB to HSV, splits the H channel, calculates color histogram, and track the face rapidly.