We use the USB camera to determine the color and ball’s position in the camera frame

• We detect balls with a Hough Circle Transform

  • Convert the frame to greyscale and apply gaussian blur to aid in circle detection

  • We get the center of the circle in the image frame (u, v)

  • We limited applied radius filters to filter out balls, other round objects, and holes

• We get the ball’s color by masking the ROI with HSV values and we select the most likely color (highest percentage)

  • We determined the HSV masking values through trial and error, with an emphasis on looser masks to make sure we always capture a ball

We first get the ball's pose in the camera frame, where we assume a pinhole model of a camera

We now want to translate from the camera frame to the base frame

Our final implementation consisted of:

1. Get ball's pose in the camera frame

2. Construct the transform between usb_camera and head_camera

3. Transform the ball to the head_camera's frame

4. Transform the ball to the base frame

We create a transform between usb camera and head

• We used both camera's to detect the same AR Tag

  • We used the LookupTransform method from TF ROS Package to get the two transforms from the usb_camera to the AR_Tag, and the head_camera to the ar_tag

  • We could then apply the equation on the left to get the transformation from the usb_camera to the head_camera

• This gives us an accurate translation between the ball's position in the camera frame to the base frame with a margin of error of +- 5mm allowing us to hit the ball

• We precomputed the translation as both cameras were static in ball detection stage, allowing for fast transformations