SOFTWARE DESIGN INNOVATION:
TO CATCH A BALL
JOSEPH MARTIN
IT TAKES A TEAM
• Building a robot requires
skills across many disciplines
• 6 weeks isn’t very much time
• Creating a robot requires
lots of experience; more
experienced members must
share their experience with
newer members
THE VISION CO-PROCESSORS
2x BeagleBone Black,
each with a 1GHz
ARM Cortex A8 CPU
USB ports for
USB cameras
Ethernet ports for
communication
with cRIO
Convenient 5v
Power Connections
Custom case to
easily mount both
BeagleBones
LOCATING THE BALL
WITH DUAL HISTOGRAM BACKPROJECTION
Original image, straight
from the camera. For the
purpose of finding the ball, it
is converted to the HSV
colorspace.
Binary image that is a
combination of figures b
and c. Highlighted in green
is the contour that is
determined to most likely
be the ball.
a
b
d
c
𝑏𝑖𝑛𝑎𝑟𝑦𝑖,𝑗
255,
=
0,
𝑡𝑎𝑟𝑔𝑒𝑡𝑖,𝑗 > 𝑘1 ∗ 𝑏𝑎𝑐𝑘𝑔𝑟𝑜𝑢𝑛𝑑𝑖,𝑗 + 𝑘2
𝑡𝑎𝑟𝑔𝑒𝑡𝑖,𝑗 ≤ 𝑘1 ∗ 𝑏𝑎𝑐𝑘𝑔𝑟𝑜𝑢𝑛𝑑𝑖,𝑗 + 𝑘2
Histogram backprojection,
based on a histogram of the
ball. Bright where colors
closely match those of the ball
from calibration images.
Histogram backprojection,
based on a histogram of the
background. Bright where
colors closely match those in
the background in calibration
images.
CATCHING THE BALL
USING CHAPMAN’S OBSERVATION
Fig 2
tan 𝜃
Fig 1
Adapted from Chapman’s 1968 paper,
Catching a Baseball
𝜃′
𝑡𝑖𝑚𝑒
B
𝜑
A
tan 𝜃 = 𝐶 ∗ 𝑡
𝑑
tan 𝜃 = 𝐶
𝑑𝑡
𝑑2
tan 𝜃 = 0
𝑑𝑡 2
tan 𝜃′ = 𝑥/𝑓
𝜃 = 𝜑 + 𝜃′
tan 𝜃 = tan(𝜑 +
𝜃′)
tan 𝜑 + tan 𝜃′
=
1 − tan 𝜑 tan 𝜃′
tan 𝜑 + 𝑥/𝑓
tan 𝜃 =
1 − (𝑥/𝑓) tan 𝜑