Host: ISCIE, SICE, JSME, JSPE, JSASS, SCEJ
Co-host: Technically Cosponsored by 48 Societies and Institutes
On ground stereo vision system is used for autonomous hovering of a quad-rotor Micro Aerial Vehicle (MAV). This kind of system has advantage to support embedded vision system for autonomous hovering and landing, since embedded system occasionally gives inaccurate distance calculation due to either vibration problem or unknown geometry of the landing target, which makes the hovering or landing process fails.
Object tracking based on weighted histogram Continously Adaptive Mean-Shift (CAMSHIFT) algorithm were used to track the MAV. The original CAMSHIFT algorithm is developed by Bradski in 1998 for face detection. However, this original CAMSHIFT algorithm fails in differentiating between the target and objects in the background that has intensity same or near to the target. A weighted histogram CAMSHIFT applies a weighting algorithm which is 1) k(r) = 1-r for pixels near to center of target (r<=1) and 2) 0 for pixels that are far from center of target (r>1). Where r is the distance of a pixel to center of target expressed by
r = {2*(x-cx)/(ws-1)}2 + {2*(y-cy)/(hs-1)}2 ,
where x, y, cx, cy, ws, hs are x coordinate in the image, y coordinate in the image, x center coordinate of target, y center coordinate of target, image width and image height, respectively. By applying this weighted histogram CAMSHIFT algorithm, the objects in the background that has intensity values near or same as target can be differentiated.
Stereo vision system used in this research is Bumblebee from Point Grey Research that comes with a function for constructing a disparity image obtained from both, right and left camera. This disparity image contains information about a slightly difference between an object position in left and right camera. By using this difference and applying a triangulation method, target position (X, Y, Z) from camera can be derived.
A linear model for quad-rotor MAV is developed based on the system identification method. Quad-rotor MAV is a small size (40cm in diameter) of helicopter with 4 rotors. The movement of quad-rotor is based on the rotational speed of each rotor.
Next, a Kalman filter based Linear Quadratic Integral (LQI) is used for controlling the MAV. The position information obtained from image processing is sent to this controller for autonomous hovering purpose.
The experimental result shows that the weighted histogram CAMSHIFT based tracking process shows a good result. Additionaly, experiment on autonomous hovering also gives a good performance.
