CV Workshop 7

MISCADA – computer vision

Workshop 7

• Using the pygame library, the following video clip can be generated

Step1: Identification

• Before being able to track the ball, we must identify it
• This mean learning its appearance
• Use (r, g) in the normalised RGB colour space

Step 1: continued

• The PNG image for the colourful ball is on Blackboard
• Read the image, create an array for the (r, g) space
• For each pixel of the ball, populate the array
• Normalise the (r, g) space to make it a probability density function (PDF)
• The PDF will be the colour model for your ball

Step 2: extracting frames

• You are given a video clip for the ball bouncing in the constraint environment
• Extract all frames and save them in a FRAMES folder

Step 3: tracking

• Using the alpha-beta tracker, track the ball throughout the video
• What you need to do is to detect where the ball is in the first frame
• The ball in the image should be the same size of the ball in the video and the frames
• So, you can use template matching to search for its first position
• You then define a bounding rectangle for the ball
• In the minimal alpha-beta tracker, you can use a fixed bounding rectangle, its position and a constant velocity