Embedded control using monocular vision: Face tracking

This paper presents a quadcopter capable of tracking a human face using an embedded system which consists of an autopilot for controlling the vehicle and an Odroid computer for image processing. Image acquisition is performed by a fisheye monocular camera. A digital treatment consisting of a Gaussian filter and a histogram equalization is applied, reducing false positives in the detection algorithm. The detection algorithm used is a cascade classifier (Haar algorithm), which is trained to detect a human face in an image. Depending on the displacement of the detected face, a Kalman filter is implemented to estimate the position and velocity which are used to follow the person.