Yes, the first screenshot of the new software is available

I used a small pendulum with the TOMI clip connected. As you can see most data looks rather good. Swing plane and Swing direction still need some attention. All angles are in degrees. Horizontal and vertical impact positions in mm, speed in m/s and swing length in cm.



Time to start looking for some beta testers...

While working on this second posting, double checking my data and comparing it against the TOMI data I found again one more problem (and solution) The TOMI assumes that the sweetspot is in the middle of the face. That is normally correct however It also assumes that the sweetspot is at the same height as the middle of the ball. That is not correct.

So the solution would be to also enter face height into the software allowing it to correctly measure impact position.

I found this when checking why my software reported a 10.5mm lifted clubhead and then measured the head. With that 10.5mm lift it would almost been a topped ball. Then I compared the head height with the ball and found that the club used has it's sweetspot around 8mm BELOW the middle of the ball.

so that 10.5 mm topped ball was now a 2.5mm sweetspot miss (below the sweetspot). Big difference!

So what could I retrieve from the data, recorded with this this 75fps@640x480 camera for which I paid Euro 25,- and Euro 5,- for the Infrared High Pass Filter. a few days programming and analyzing around 5gigabytes of data.



It was fun doing this!

A owner of a TOMI system requested a responds from Pure Motion, the makers of the TOMI.

This is what he received:

The author makes a critical error in” assuming” (his word) that the TOMI measures the starting point of the putt as an average of various positions during the 10 second calibration countdown. This is not the case. The TOMI even allows an instantaneous calibration method of zero seconds! Using this false assumption as a basis going forward to analyze the TOMI’s accuracy puts all of his results in question. That being said however, one must realize that neither the TOMI nor any other current system available, even those costing 10 times as much, can make all of its measurements accurately down to the millimeter. I would refer those who have questions regarding this topic to the independent McKenzie study published in the Journal of Sports Sciences 2010. Linked here: http://www.ncbi.nlm.nih.gov/pubmed/20509092. 


Regards,

Rick Gorbell
Director of Operations
Pure Motion, Inc.
Cell: 512 289-4085
rick@tomi.com

It can be found here : http://forum.ottawagolf.com/showthread.php?74195-Trackman-vs-foresight-gc2-with-hmt&p=496828#post496828

As I said in that forum I must thank him for the reply as it provides a link to a research paper with some more insight into the TOMI system. But his reply holds no ground and doesn't address at all any of the reported issues! He tries to claim that the assumption is wrong and therefor all is wrong. However I think he has to read the study himself. Because I just did and guess what ....it confirms all my assumption!

A small quote from the study

...The system captures the coordinates of the LEDS in the calibrated position...

However reading that paper in more detail it's clear that they did not use the values reported by the TOMI software in this report. They made an adjustment in the TomiPCPro.ini allowing them to get the raw data from the TOMI.

This is a part from the .ini file

;Turn on or off capture video during a set 0= nocapture 1=capture video
 CaptureVideo=0
 ;Turn on or off writing CSV files 0= no file created 1= files created
 WriteCSV=0

When setting the CSV flag the TOMI software will generate a CSV file for each recorded stroke. This is a copy from a CSV file that I generated.

Frame,time,time2,dtime,px1,py1,px2,py2,px3,py3,px4,py4,x1,y1,z1,x2,y2,z2,x3,y3,z3,x4,y4,z4,gx,gy,gz,a1,a2,a3,da1,da2,da3
-1,,, , ,, ,, ,, ,,  -0.114373,6.24025,2.03827, -0.188196,10.2681,3.35389, 2.21283,8.36562,4.19439, -2.40952,8.22516,4.34292, -0.538337,0.712824,31.6788,  -1.58325,-0.952291,17.9052
0,,, , ,, ,, ,, ,,  -0.0414328,6.23576,2.05476, -0.0682496,10.2718,3.38467, 2.31798,8.35873,4.20283, -2.28448,8.21643,4.42285, -0.0788204,-0.0139887,-0.218871,  -1.47711,-0.282207,19.0152
.
.
.
30, 333.14501,333.16786, 0.03613,  -38.1869,-50.5, -38.0065,31.4725, 14.3937,-11.5, -80.8201,-12.5, -0.373691,6.34273,1.9672, -0.269456,10.3369,3.2686, 2.10078,8.4227,3.96498, -2.49171,8.33695,4.33034, -0.536554,0.10196,0.0245209,  -5.11606,1.49129,18.0318,  -2.0557,1.7735,-0.983453
.
.
39, 333.49647,333.49302, 0.03613,  138.5,-37.5, 113.5,40.5, 164.133,7.5, 75.5,-11.5, 7.22171,6.92283,2.19972, 6.46002,10.8678,3.48074, 9.05709,9.44811,4.49454, 4.5907,8.43018,4.50394, 8.40759,0.780812,0.373935,  -0.690968,-10.9259,17.6921,  2.3694,-10.6437,-1.32314
Impact,dx,dy,dz,x,y,z,vel,angV,angH,dxAddr,dyAddr,dzAddr,da1addr,da2addr,da3addr,d1g,tstart,timpact,tback,tbstrk,tfwstrk
30,  1.13247,-0.00959,0.00506,  0.00000,-0.01265,0.06730,  31.35,0.19,1.24,  -0.08,-0.01,-0.22,  -1.48,0.67,1.11,  6.56570,  332.120,333.185,332.770,0.650,0.415

The content of this file confirms that the frame rate is indeed 30fps. It also confirms that it uses the calibrated point to "find" the impact moment. In this CSV file you can see from the last line that the impact is around frame 30.

The researcher then used that raw data from the CSV file (or better the position data from the four leds) and then used THEIR OWN software to calculate the results!! So the results and conclusion you read in the paper are only a report how good the leds position in the software where translated to points! It doesn't make any comments about how the values that the researchers calculated compares to the values the TOMI software reports.

The good thing about this research paper is however that it explains that the design of the marker unit (or clip) is good enough to be used for a putter swing!

For those that want to read the research paper themselves : http://www.yumpu.com/en/document/view/15763669/validity-and-reliability-of-a-new-method-for-measuring-putting-stroke-

PS Forgot to mention one other find in the ini file

;
; Number of non-moving frames required to be at address
AT_ADDRESS_FRAMES=10

; Number of non-moving frames required to calibrate.
AT_CALIBRATE_FRAMES=10

; Number of frames to examine for swing start motion.
SWING_MOVE_FRAMES=10

So that "instantaneous calibration" claimed by Mr. Corbell must be at least 10 frames long......

I believe that it's important for a clubfitter to understand his tools. This posting is about the TOMI, how it works, what the data it produces really means and how that data can be improved. There is a lot to tell about it and some of that information is very technical. I’ll try not to be too technical in this first posting.

Summary of the results

The TOMI system can’t measure the moment of impact. It can’t measure the correct sweetspot or clubhead orientation and direction. Because of these two issues the information it produces has large error margins.

Description of the TOMI system

The TOMI consists of three parts, A PC webcam
According to the driver that came with the TOMI software it’s based on a VIMICRO ZC0301PLH. The VIMIRCO website (VIMICRO) states it’s a USB 1.1 compatible webcam capable of streaming up to 30fps at VGA resolution. This webcam became available around 2005 and can support different CMOS image sensors. As I did not open it up it is unknown which CMOS chip it is using and at what resolution and speed (fps or frames per second) it is actually recording the swing. The video option in the TOMI software suggests it is using a 320x240 resolution when measuring the size of that video window.

The second part of the system is the marker unit or clip (transmitter)
which holds four infra-red leds. The arrangements of the leds are such that when an image (2D) is made from them it is possible to determine the 3D Pose of the marker unit. More background information can be found here http://en.wikipedia.org/wiki/3D_pose_estimation.

The third part is the TOMI software
I used the latest version (1.33) which was released around January 2008. This software is responsible for the calculations based on the images provides by the webcam. Those images contain only the four leds

Determining the 3D pose using the TOMI transmitter.

The Marker

It important to understand that this marker technique that the TOMI (and other systems) use will only measure the position and rotation of the marker, it doesn’t measure the clubhead or sweetspot. Only the marker. So it is possible to keep the head at the same position and move the rest of the club and thus the marker

, or move the head and keeping the marker at the same spot.

However as the axis of the shaft is parallel* to the marker and connected to the marker at a fixed and know distance we could use some math to determine were the tip of the shaft is when the position of the marker has been moved and/or rotated.

The manual states “Place the clip on the shaft with its base approximately 10 inches from the ground.”

Is that measured vertical from the ground or measured along the shaft? 10 inches from the top of the base or the bottom? And why “approximately”? If we do not place it at the same spot each and every time it’s obvious that that doesn't help to determine where the tip of the shaft is, especially because the system has no idea where the ground level is as it only sees the marker.

Because of these unknowns we have no idea which point the TOMI is actually calculating as the tip of the shaft.

Is it the green dot, the blue or the Red? Also note that all thee point are NOT in the sweetspot! What about a center shaft design?

Here we also followed the axis of the shaft. And if we would use the correct length in the calculation the tip of the shaft is the exactly in the sweetspot.

What about a design with an offset?
If we follow the axis of the shaft we are in front(!) of the clubhead.

Please remember these three examples while reading the following part.

The TOMI system doesn’t know when the impact really happened. It can’t see anything else then the four leds. It might be that the clubhead speed at impact slows down enough to clearly show the impact moment but I have included a speed chart at the end of this documents to show that impact doesn’t slow down the clubhead significant enough for the TOMI to use speed as an impact indication.

That leaves only one option for the TOMI to determine impact position and that is “calibration position”!

The following is what I assume it is doing but please note that there is no other option available for the TOMI system!

During the calibration procedure it measures\calculates a position, this position is possibly related to the 10 inch marker placement instruction. So during a 10 seconds calibration and with 30fps it gets 300 positions. When the calibration is finished it then calculates the average for the calibration point. This is a 3D point, it knows the position on the x-axis, height (y-axis) and depth (z-axis). It also knows the rotation of the marker at this calibration point. It will then assume that this point is also very close to a possible impact point. If it records a new position during the swing that is further to the right (on the x-axis) then the calibration point it can assume that that point crossed this virtual "finish-line". So when during the swing any data point is measured as passing the finish line then between that datapoint and the previous one the impact must have been taken place.

Imagine that the TOMI now indeed knows some virtual shaft tip position

The big yellow circle is the golf ball. The red line is the finish line, the black dot is the tip of the shaft for which its position is calculated by the TOMI software. The green line is the topline of the clubhead.

During the calibration period we place the sweetspot as close as possible to the ball and we cross the finish line at the exact same spot but with an open face and with the same three different clubhead designs as noted above.

In scenario A we have a centered shaft club head and at the moment we return to the calibration point we hit the ball with an open face. The system could report a center hit with an open face which is correct. However in scenario B we have also crossed the finish line with the calculated shaft tip but because of the design being a non-centered shaft only the tip of the shaft has crossed the finish line and there is still time for the golfer to close the face before the sweetspot will actually hit the ball.

Scenario C is the same as scenario B but now because of the extra offset there is even more time to close the face; it might be even possible to get it closed before really hitting the ball. If the offset would have been backwards the impact already happened before the virtual shaft tip crossed the finish line.

In both scenario B and C it is very unlikely that the reported face angle by the TOMI software is correct. The design of the head is important!

If you still with me till this point then remember the remarks about the height of the placement of the marker on the shaft and the trouble of finding the correct end of the shaft. Here are again three scenarios.

In these three scenarios we have three different “heights” for the shaft tip. Green is when the tip is measured too high. Blue when it’s correct and Red when too low.

Yellow is the ball, the red line is the finish line and the thick brown line is the shaft.

When hitting the ball with no shaft lean (scenario A) there is no problem with the error in the height. However in the B scenario’s with forward shaft lean we see that if the shaft tip height is calculated too high (B) we did not yet hit the ball and in scenario B II we already hit the ball before the TOMI assumed the crossing of the finish line!

The C Scenario with backward shaft lean show exactly the opposite. Too high (C) means we already hit the ball before the TOMI thinks we crossed the finish line and C II we did not hit the ball yet.

Combine the above scenarios with a calibration where you did not had the shaft in a 100% correct vertical position and all bets are off when and where the ball is when the calculated virtual shaft tip crossed the finish line!

In both scenario B and C it is very unlikely that the shaft angle at impact reported by the TOMI software is correct.


Conclusion: Head design determines what virtual position we are really measuring / calculating. As head design data can’t be entered into the TOMI system it’s unknown to the TOMI system where the sweetspot is. It can therefore not measure clubhead orientation or position.

Solution :



If we could enter the following information
A : The distance from the top led to the horizontal sweetspot line.
B : The distance from the sweetspot to the shaft axis line.
C : The LIE.
D : The offset, or distance from the sweetspot to the shaft axis.

Then that data could be used to calculate exactly where the sweetspot is in relation to the marker.

NOTE I: During calibration the ball must be placed as close to the sweetspot as possible, for every shot the ball should be placed on exact the same spot as it was during calibration.

NOTE II: The shaft must be held still and without any shaft lean during the complete calibration period.

I used the word “parallel” a few sentences back. But is the marker really parallel to the shaft? It is if you use a non-tappered part of the shaft to connect the marker on. But most shafts are tapered and that means that the marker is not parallel to the axis of the shaft.

Black is the tapered shaft en blue is the marker.

The error is not that big and depends on the size of the steps. For the putter I used in the tests the difference in step size was around 0.3mm so an error of 0.3 degrees which means we would miss the middle of the shaft at the tip with an error of around 1.5mm.

However there is a second issue with the shaft. Not all shafts have the same outside diameter. And it is important to know the exact position of the shaft rotation.

Above the image of the marker rotated from right to left on a vertical oriented shaft. As you can see the rotation is visible but the images also shows some displacement. The led on the top in the left image was measured as being on the x-axis position at 531.7mm and in the right image at 550.5 so a movement of 18.8mm where in reality it was not moved only rotated.
So without the correct information about the shaft diameter the software would report the club as being moved and rotated (red lines) and not as only being rotated (blue lines).


And lets not talk about the fact that this error also will influence speed calculations and might report the club as going across the finish line were in reality it still might not!

Conclusion: The TOMI system will report head displacement where it could be only rotated because the shaft information is not entered into the software. Also the tapered shape of the shaft introduces a (minor) error.

Solution: Shaft diameters at both positions where the marker is connected to the shaft should be entered into the software to overcome this issue.

I could continue explaining the consequences of the camera placement for the data or camera calibration for lens distortion or the frame-rate and resolution for the quality of the data or explaining the 4 inch impact zone but let me finish this first posting about the TOMI with some data.

4 swings were recorded with the TOMI and a high-speed camera (850nm Band Pass Filter) running at the same time. That camera was running at 75fps@640x480. With that camera also some calibration data was recorded.

During the calibration run a special rig was used to place the marker unit in a vertical position, laser aligned with the target.

The reported position relative to the camera is in mm, rotation in degrees. First values are the averages for 247 frames and the second line are the standard deviation values.

This shows good rather good precision for a marker based system using such a camera!

The following is the displacement recorded (mm) between each frame. Impact is around the 149 mark. Please note the rather stable data before the swing really begins. No "smoothing" filter was used on this data. It's the raw data


Below is the face angle rotation for the same swing.


And a close-up of face angle data just before and after impact.


When the sweetspot is at the impact position for this 240cm put it is moving around 18mm between frames. With the assumed 30fps for the TOMI the sweetspot for the TOMI would be moving at 45mm between frames!

With the face at impact closed -2.4degrees and closing more then 1degree between frames then again for those 30fps with the TOMI the face would be closing more then 2.5degree between frames!

In the 5 frames around the impact (2 frames before and 2 frames after) the Rate of Closure for this swing peaks at 102degrees/sec!

The first swing in the TOMI screenshot below is the same swing as shown above.

I recorded an alignment at address of -0.1 (standard deviation 0.1)

Dave Honke suggested to measure the shaft again after its FLO was found.

CPM @ FLO : 391.07 ± 1.33
CPM @ FLOW +90° : 389,61 ± 0.65
CPM @ FLOW +180° : 390,92 ± 0.80
CPM @ FLOW +270° : 390,11 ± 0.99

As you can see below the cpm is still increasing.

In order to understand why that standard deviation was so large when using the raw data we must understand how the frequency is measure\calculated.

Frequency is the number of periods per unit time (per second) and is typically measured in hertz. By multiplying this value by 60 we get the CPM (cycle per minute).

The period is the time for one complete cycle of an oscillation of a wave and can be measured by taken any two corresponding points on a waveform and measure the time-interval between these points.


I decided to use the tops and the bottoms of the oscillations recorded as these are very clear points in the data and it's easy to find the corresponding points. The corresponding point would be the next top or bottom of the next oscillation.

Looking at the recorded wave those points look rather sharp and easy to identify.


However here is a close-up:


It's easy to see that there is no real simple top or bottom to be determined nor does it seem easy to find the correct corresponding point. That explains the rather large standard deviation found, the software will find some tops or bottoms (maximums or minimums) but those might not be the correct ones.

Next step:

So the next step was to implement a software routine that could remove some of that noise from the data and improve the shape of the waves. This is the result after applying a software routine for doing just that:


And here are all the CPM's found using the now filtered data:


That looks much better!

Using the tops of the oscillations the average CPM found is now 392.02cpm with a standard deviation of 0.84cpm. Using the bottom of the oscillation it's also 392.02CPM with a standard deviation of 0.84cpm.

But the big question that remains is : Why does the measured CPM increases over time? From around 391cpm to 394cpm?

Since I started clubfitting I use an Auditor Frequency Analyzer. This model uses strain gauges to measure the frequency and is capable to show the cpm up to 1000 cpm on it's LCD display.

However this model also has something very nice, it has an RS-232 port and analog out!


Using the analog out would not only allow to measure cpm's higher then the 1000 cpm LCD limit but would also allow to measure shaft specifications like decay rate and amplitude. The ports are blanked out. Please note : By breaking the tabs you signify your intent to use the extra features as is and at your own risk!

So I broke out those tabs to get access to the analog out port, found a audio cable and contected the cable to the "Microphone in" port on my laptop. Wrote some software and this is what I found.


This is a True Temper Dynamic Gold X100 and I was using a 205.7grm tip weight. While I recorded the analog data I also used the Analyzer to give me the CPM's. These are the CPM's displayed on the LCD output:

391.13
391.90
392.15
392.92
394.99

The average according to the analyzer was 392.61cpm.

I then took the analog data (all 38 seconds), did not clean it but just tried to extract the T (period) values from it.


As you can see there is some serious spread in the values. In fact the average is 392.18cpm which is very close to the reading of the Analyzer itself but the standard deviation is 9.49cpm!

to be continued.....

Some years ago (feb-2010) I posted the results of my undertaking in understanding the Zelocity PureLaunch. In that paper I included the link to the test software allowing any owner of this LM to see if he could duplicate my results. So how many did that and shared the results with the community?

None

Because I thought that there could still be a way around the issues with the Zelocity I decided to create some new (free) software with improved math and algorithms. So after a few month programming I released it and also created a forum for those using it. The forum could be used to ask questions, share ideas, post improvements and share results. So how many did downloaded the software, used it and shared their experiences?

12 members are registered on the forum and 8 members posted 31 messages on the forum. So 31 messsages in 20 months for a software product that improves the LM results used by a few hundred clubfitters.

For now it seems that the "clubfitter community" doesn't exists when talking about Launch Monitors and their results

Now my problem is I have a second "understanding the ....." paper in the works and again I will include software allowing using that machine but with better and improved results. Will the "community" wake up and join? I don't know but we will see...