Abstract:
Bubble detection and tracking is an important and complex part of motion analysis
of two phase air-water flow in non linear wavy channels in which bubble speed
and size varies and sometimes bubble occludes with the boundaries of the other
bubbles. Due to optical noise bubbles boundaries are not recorded completely in a
camera and then need to be filled intelligently to recognize them as single entity in
a non-convergent sinusoidal channel. To solve this problem non linear sinusoidal
curved channel is transformed to a linear straightened horizontal channel and
then trail and error method of segmented morphological operations are applied to
identify each bubble as a single entity and should not be occluded with the other
bubbles. The detected bubbles are associated across different frames based on
motion, estimated by Kalman filter to predict frame by frame detections and the
likelihood of every detections assigned to each track. Since a Kalman filter is a
good estimator for linear motion so path based approach to the sinusoidal channel
is more robust, as it linearizes the motion of the sinusoidal channel to study a non
linear motion model. Finally the result shown that our path based straightened
channel gives more accurate results for both bubble count and computational
velocities than non-convergent sinusoidal channel.
