Stereoscopic vision camera software for your CX camera systems
As an alternative to the standard X,Y camera setup for CX camera systems, the new stereoscopic image processing software will allow you to mount both ball tracking cameras on the ceiling instead of one on the ceiling and the other on the floor.
This software development is being done as some customers would prefer not to have any cameras mounted on the floor.
The disadvantage with the stereoscopic method, however, is that the user can no longer see the exact ball launch angle and path in the captured images as the traces in the images will appear skewed.
Comparing the stereoscopic camera setup to the GSA Golf standard X,Y,Z setup
The above two images show the camera images captured using the standard X,Y,Z camera setup.
The left image is from the floor side mounted Vcam and clearly shows the vertical launch angle of the shot. The length of the ball trace is directly proportional to the speed of the ball (Z axis).
The right image is from the overhead mounted Hcam and clearly shows the ball's direction left or right.
In contrast, the left and right stereoscopic camera images do not show vertical launch or ball direction in any direct user readable form.
Instead, the image processing software has to calculate these values using rather complex mathematical algorithms.
Note: As GSA Golf's product tradition has always been to allow the user to self verify the data being captured by the system,
this method goes against this philosophy and - unless it's absolutely essential that all cameras be overhead mounted - we are not recommending stereoscopic setups.
However, as there has been some interest in using this setup we have introduced it.
Stereoscopic Setup procedure
The setup procedure for stereoscopic vision cameras is somewhat more involved than the traditional XY setup and may also require some calibration.
Mount the V and H cams around 4 to 5 feet apart either side of the center line.. The Vcam is on the right hand side and the Hcam on the left.
Ensure that the camera orientation is such that the screen is on the left of the image (select X reverse if not) and that the camera is aimed at the center line of the enclosure
The IR lamp and SX line scan camera remain as before in the center.
In the CP's Setup window select "Stereoscopic".
Important: Ensure you are using CP version V 18.104.22.168 or higher and that you first "right" mouse click on the "Defaults" button
Aligning the cameras
Place a golf ball directly under the IR lamp on the center line. Note: you can also use a white poll that extends from the ball launch position to the center line.
This will represent a ball trace nicely and may be easier to align the cameras so that the image is square to the screen.
Using the "Video stream mode" aim camera so the ball appears in the center of the image.
Select the "Ball launch position X,Y" button and using the up/down, left/right keyboard arrow keys move the green cross hairs to the ball launch position. Ensure that this position is exactly in line with the center line.
Do the same procedure for the left camera (camera 2).
The systems is now ready to run but may require some calibration.
Calibrating the stereoscopic cameras
Stereoscopic camera calibration
Stereoscopic vision camera software for your GSA Golf CX camera systems will be available soon
Release date for this free software update is scheduled for the end of April 2017
As an alternative to the standard X,Y camera setup for CX camera systems,
the new stereoscopic image processing software will allow you to mount both ball tracking cameras on the ceiling instead of one on the ceiling and the other on the floor.
Note that there will be a standard default calibration for stereoscopic camera setups that won't require any user adjustment for basic play
Two 3d co-ordinates of the ball trace are captured with stereoscopic camera setups
Formulas are used to determine speed, direction and launch angle of the ball
Which is better: XY or Stereoscopic?
In regards to ball tracking in a golf simulator, at present there's still a debate as to which method is more accurate.
Certainly XY setups are as accurate - if not more so - than stereoscopic setups providing there is sufficient calibration.
Our current tests actually show that there are issues with short spacing parallel stereoscopic setups with low shots.
i.e. it's ability to differentiate between a low shot and a ball skimming or rolling along the ground
whereas XY setups have no problem with this.
Also a major drawback with stereoscopic setups is the fact that the user cannot easily see the vertical and horizontal launch data simply by looking at the captured images.
In contrast to XY setups - that clearly show vertical ball launch angle and ball path in the captured images - the camera images in a stereo setup won't make much sense to the average user.
Aligned and angled camera positions
In contrast to the parallel stereo camera setup where the cameras are only spaced at around 1 ft from each other and point straight down,
current tests are showing far better results when the cameras are spaced at greater distances (6ft or more) and angled to and aligned to a base center point as in the above image.
Note that there will be a standard default calibration for fixed camera spacing and mounting heights so that no additional calibration will be required by the user.
The calibration method consists of placing a large chequered image printout mounted on a board directly under the cameras at various heights.
Each square on the board represents a golf ball position at various positions in 3D space.
Note that if we didn't use a chequered board, we'd have to place hundreds of golf balls under the cameras at precise positions - a near impossible task.
The board is first placed at ground level and the cameras are triggered (via a soft trigger in the CP) to capture the image in both cameras.
Each square's XY position in both left and right cameras is recorded together with the real 3D XYZ position of each square.
The procedure is repeated 20 times while raising the board 10 cm upwards with each step for each area segment.
Total height is then 200 cm ( 6.5ft)
Note that the size of the chequered board has to be entered into the CP's stereo calibration window before this procedure is started
together with the real 3D position of the board itself.
The number squares on the board is automatically detected but if insufficient squares are available for the entire scan area
(usually 4ft X 4ft ) then multiple scans will have to be made to cover this area.
Precise and level method of raising the chequered board under the cameras during the calibration process
The calibration creates a 3D matrix that converts XY ball pixel positions in the camera's 2D images into real 3D XYZ positions
These real XYZ co-ordinates are then used to calculate precise ball launch angle, speed and path
This stereo calibration feature will be available in the CP within the next 2 weeks or earlier
Note that there will be a standard default calibration for fixed camera spacing and mounting heights so that no additional calibration will be required.
Stereoscopic image processing explained
Up until now our camera ball tracking cameras are mounted in an X,Y configuration.
That is, one camera is floor mounted facing the ball trajectory from the side to see the launch angle (Y) and the other over head mounted facing down to see the ball path (X).
An alternative method of measuring these two angles is to mount both cameras in the upper left and right corners of the enclosure and use what is known as stereoscopic image processing.
You may have noticed that a number of high-end golf simulators use this configuration and method and we will be offering this too soon.
The following is an explanation of stereoscopic image processing relative to a golf simulator.
The white lines in these images represent ball traces and angles are measured from a known launch position and center line.
The images and angles appear to be exactly vertically flipped in this case.
It is only when comparing the right camera image to the left camera image do we see what these true angles are.
The first image above is of a ball rolling left on the ground at 20 degrees left. In the right camera this image appears vertically flipped.
If the angles are exactly reversed then we know the ball had no vertical launch angle and just rolled.
Here, the ball is launched with a 20 degree launch angle with zero degree path. i.e. dead straight. The right camera sees an exact horizontally flipped image of the ball trace.
If the angles are reversed exactly 180 degrees then we know that there was no path component and only vertical launch angle.
Here we have a mixture of ball path and launch angle. The direction component (left in this case) draws the ball trace further towards the center line in the view.
If we measure the angles we can determine the ratio of ball path and launch angle in the shot. So in this case 25% of the angle is the ball path component and 75% the ball's launch angle.
The angles are scaled to real world path and launch angles depending on the camera's mounting angle.