Depth Information Learning For Face Features Extraction And Detection

Augmented Reality technology development is very rapid, people as an important member of human-computer interaction in augmented reality system, research of human body, face, hand detection and identification has become hot and focus. Face detection is one of the key technologies of information processing in the human face, is the basis for other research techniques about face. Based on the position and orientation information provided by the face detection, face calibration, identification, tracking and even facial recognition and body gesture tracking can be depth study. So the performance of face detection system is good or bad is very important. How to quickly and accurately detect faces and try to avoid the occurrence of false detect is quite a complex and difficult task. Most of the current face detection technology is based on planar images, as for the non-real face on the photos and clothes, they have the same characteristics as real human face, easily caused accidental detection. So, how to reduce this type of error detection meanwhile ensuring the detection speed and accuracy have great practical value and significance.The main content of the paper is the face detection based on statistics. Our study started from the face depth information, using the latest binocular depth camera to get the color images and depth images of face. In the image pre-processing, using gray color space conversion and Gamma normalized color images; to deal with the raw depth information, using UV mapping and image smoothing methods. In image feature extraction, design and implementation of an improved HOG feature combined LBP feature to characterize the face color and depth features. It has gray-invariance, rotational invariance and reduce the effect of the light and shadows. And The use of machine learning SVM method in the offline is a two-stage classification training system, to training a good detection rate, low false detection rate and robust classification of2D and3D human face classifier. Finally, we developed an online video detection system, in the detection, first filtered using2D classifier, and3D classifier mean to exclude false detection, the methods can ensure the detection speed and accuracy, while greatly reduced the non-real face caused false detection occurs. It verified the effectiveness of the proposed method in the paper.