A NOVEL APPROACH FOR FACE TO CAMERA DISTANCE ESTIMATION BY MONOCULAR VISION

To accomplish the face to camera distance estimation, a method based on monocular vision by using a single camera is proposed. There are three key steps to complete the estimation: extracting and locating the feature region in face, calculating pixel area of characteristic triangle, and constructing the measurement formula. To detect and locate the feature region fast, the traditional AdaBoost algorithm is improved by reducing the features' quantities and expanding samples. The measurement formula which is revealed the relationship between pixel area and distance is derived from the pinhole camera model, camera calibration and area mapping. The measurement formula can be constructed dynamically during the system initialization. The optimum moving range for measurement formula constructing is also defined. After analyzing experimentation, the accuracy of measurement is above 95% and the time each measurement cost is about 230 ms which satisfied the precision and real-time requirement.