A Closer Look At The AimTrak (with video goodness)

[Ginsu Victim]

Speaking of PC applications, any word on a new version of the AimTrak software?

[RandyT]

Yes. If larger numbers of LEDs were used, they would need to be controlled by the gun logic so the gun would know which LED(s) it is looking at. This means the simplicity of the design is lost because a controller board would be needed and/or a PC application to control everything, plus more wiring, driver circuitry etc.

No, that's not entirely accurate.  It could also simply reverse the orientation.  To use the Aimtrak as an example, it has a row of LEDs forming a line to left and a single one on the right.  If one were to add another bar, it could have 3 evenly spaced LED's, or the line of LED's to the right instead of the left.  All of this can still be handled in firmware, without any intelligence in the lighting.

Of course, it really isn't important to know which is which, unless one needed to fire off-screen to the bottom or top.  It would simply see the LED's below the calibrated centerline and provide coordinate information as though it were the other bar.  This would be a small tradeoff if it allowed one to get closer to the screen, as a left and right "offscreen reload shot" is still possible.

RandyT

[RayB]

Does this device see "rotation" too?

[KissMyWookie]

Do you think that perhaps the LEDs in arcades also blink at different frequencies so that if the camera can only see one it know which one it is based on timing?  Sort of like a lightgun and CRT?  or maybe some of the "lightguns" in arcades are ultrasonic sort of like the Power Glove...

Yes. If larger numbers of LEDs were used, they would need to be controlled by the gun logic so the gun would know which LED(s) it is looking at. This means the simplicity of the design is lost because a controller board would be needed and/or a PC application to control everything, plus more wiring, driver circuitry etc.

If there were 4 IR emitter boards, each identical and "L" shaped. They could be placed around each corner of the monitor and the gun firmware would know exactly which corner it was looking at by the orientation of the "L".

If it was done this way, distance would no longer be a factor providing at least one corner was visible. Auto calibration and linearity would be possible too.

Of course, this is too late to implement now, but might be useful in the future.

Steve

[brandon]

Do you think that perhaps the LEDs in arcades also blink at different frequencies so that if the camera can only see one it know which one it is based on timing?  Sort of like a lightgun and CRT?  or maybe some of the "lightguns" in arcades are ultrasonic sort of like the Power Glove...

Yes. If larger numbers of LEDs were used, they would need to be controlled by the gun logic so the gun would know which LED(s) it is looking at. This means the simplicity of the design is lost because a controller board would be needed and/or a PC application to control everything, plus more wiring, driver circuitry etc.

If there were 4 IR emitter boards, each identical and "L" shaped. They could be placed around each corner of the monitor and the gun firmware would know exactly which corner it was looking at by the orientation of the "L".

If it was done this way, distance would no longer be a factor providing at least one corner was visible. Auto calibration and linearity would be possible too.

Of course, this is too late to implement now, but might be useful in the future.

Steve

wow.. that makes way too much sense! why didn't I think of that :)how about for the ultimate add a second sensor at the back of the gun either near the "hammer" or under the "clip"  Then you could pivot the gun in place or move in perpendicular to the screen and it would always know its angle and position in 3D space buy comparing the data from two points in space. Thats getting overly complicated I know

[KissMyWookie]

wow.. that makes way too much sense! why didn't I think of that :)how about for the ultimate add a second sensor at the back of the gun either near the "hammer" or under the "clip"  Then you could pivot the gun in place or move in perpendicular to the screen and it would always know its angle and position in 3D space buy comparing the data from two points in space. Thats getting overly complicated I know

My point was that it is possible to match the ease of use (distance/calibration/accuracy) of the newer arcade lightguns, without having special driver circuitry or complex synchronization between each corner emitter. The emitters would be statically lit IR LED's, with only power going to them and the aiming calculations would be performed "in gun", as they are on the Aimtrak now.

[Ginsu Victim]

If there were 4 IR emitter boards, each identical and "L" shaped. They could be placed around each corner of the monitor and the gun firmware would know exactly which corner it was looking at by the orientation of the "L".

Damn, that sounds like it might work. I don't know the technical side of it, but it does make some sort of sense.

[brandon]

If there were 4 IR emitter boards, each identical and "L" shaped. They could be placed around each corner of the monitor and the gun firmware would know exactly which corner it was looking at by the orientation of the "L".

Damn, that sounds like it might work. I don't know the technical side of it, but it does make some sort of sense.

unless you tilt the gun "gangsta style" lol  Then your upper left corner becomes your upper right, etc. Personally I never do that so...  I guess you could add a gyro sensor

[Neverending Project]

Does this device see "rotation" too?

By "rotation" do you mean if you were to extend a line out of the gun barrel and pivot the gun around that line? If so, then yes, to a small degree. There is a setting in the AimTrak software to compensate for this called Tilt/Z Correction. You can turn it off (smooth mouse movement but goes uncalibrated when you tilt the gun), Intelligent (it applies correction intermittently, so if you hide the cursor you won't notice) or Continuous (jittery mouse cursor, but continuously applied correction).

I played with it a little, and it certainly makes a difference, although it does not allow you to use the gun sideways (gangsta style?), ,nor would I recommend tilting the gun excessively in any normal gameplay if you want to keep where you aim. If you use the crosshair, then it may be fine.

[AndyWarne]

If there were 4 IR emitter boards, each identical and "L" shaped. They could be placed around each corner of the monitor and the gun firmware would know exactly which corner it was looking at by the orientation of the "L".

Damn, that sounds like it might work. I don't know the technical side of it, but it does make some sort of sense.

The problem with that is if the gun were able to see all four corners that would be simply too many LEDs for the image processing logic to handle. The problem with these kinds of approaches is the gun has to be accurate whether it can see one, two, three or four of the groups of LEDs.

[Ginsu Victim]

See, I told you I didn't know the technical side.

[AndyWarne]

Of course, it really isn't important to know which is which, unless one needed to fire off-screen to the bottom or top. 

It is necessary to know which one its looking at. It is a pre-requisite that the gun needs to work even if it can only see one of the groups of LEDs, otherwise the field of view of the device is greatly reduced, which means you have to be much further away. This is the problem with the Topgun.
Actually I have tried something like this as part of the development, having two LED bars, and making an educated guess which one it could see, if it could see only one. There are two problems, firstly it gets the guess wrong in certain movements such as moving off screen, around the edge and back on screen again, and also there is a jump when it transitions from referring to the first, both, or second bar, which cant be fully eliminated.
I spent 9 months researching the raw sensing solution and I believe I covered all options.
If the min distance limitation is taken out of the requirements, all kinds of options are possible.

[bkenobi]

I'm intrigued by the wide angle lens mod for the Topgun.  Would something like that work for this camera?  Most guncons that people are looking at have a lens to start with, so would using that (or more likely a better lens) to see more of the screen from closer up help?  I know it would distort the camera's image so a firmware upgrade might be needed (or not if the calibration somehow worked like in the case of the Topgun).

[RandyT]

*edit*  I'm going to pull back on this until I can flesh this out some more with some diagrams.  I think I see what the issue is based on the way this is designed.

RandyT

[KissMyWookie]

Of course, it really isn't important to know which is which, unless one needed to fire off-screen to the bottom or top. 

It is necessary to know which one its looking at. It is a pre-requisite that the gun needs to work even if it can only see one of the groups of LEDs, otherwise the field of view of the device is greatly reduced, which means you have to be much further away. This is the problem with the Topgun.
Actually I have tried something like this as part of the development, having two LED bars, and making an educated guess which one it could see, if it could see only one. There are two problems, firstly it gets the guess wrong in certain movements such as moving off screen, around the edge and back on screen again, and also there is a jump when it transitions from referring to the first, both, or second bar, which cant be fully eliminated.
I spent 9 months researching the raw sensing solution and I believe I covered all options.
If the min distance limitation is taken out of the requirements, all kinds of options are possible.

Sorry, I assumed that with 4 "L" shaped emitters, the LEDS's on each corner emitter would be a fixed distance apart. The gun logic would calculate the pointer position based on the distance between the LED's on the visible corner and not the distance between all 4 corners... think of it as enhancing the way the Aimtrak currently works by adding extra emitters (and so letting the position be calculated if the top of the screen is not necessarily visible, but can work with just one corner visible).

[RandyT]

I took down my earlier post, because I assumed after the fact that the CMOS sensor was unable to ever see two emitter boards at once, unless it was a good distance from the screen.  When aimed at the center of the screen, the CMOS sensor would absolutely need to have more than one set of emitters in frame for any of the advanced stuff related to multiple emitters to be done.

I think there's not enough information about the hardware available, specifically the field-of-view of the CMOS sensor / lens, to know whether this system can be improved at all.  Andy mentioned something about a cell phone camera earlier, so I did some testing with my (t)rusty old Motorola Q.  One thing that is very interesting is that I can get the entire monitor in frame, with some extra space above and below for emitters from about 27" in front of the emitters.  This is about 2' closer than where I need to use the Aimtrak for it to work well, so I am going to have to make the assumption that the camera on the Q cell phone has a bit of a wide angle lens.  I suspect that such a lens would be very undesirable on the Aimtrak, due to the distortion inherent with inexpensive wide angle lenses, so it probably has a much narrower field of view, which provides a more uniform image.

If this isn't the case, however, I'm having a hard time understanding why a second set of emitters, with a reversed pattern (i.e. same board, flipped) or not, wouldn't allow one to get quite a bit closer than is currently possible.  

RandyT

[AndyWarne]

I took down my earlier post, because I assumed after the fact that the CMOS sensor was unable to ever see two emitter boards at once, unless it was a good distance from the screen.  When aimed at the center of the screen, the CMOS sensor would absolutely need to have more than one set of emitters in frame for any of the advanced stuff related to multiple emitters to be done.

I think there's not enough information about the hardware available, specifically the field-of-view of the CMOS sensor / lens, to know whether this system can be improved at all.  Andy mentioned something about a cell phone camera earlier, so I did some testing with my (t)rusty old Motorola Q.  One thing that is very interesting is that I can get the entire monitor in frame, with some extra space above and below for emitters from about 27" in front of the emitters.  This is about 2' closer than where I need to use the Aimtrak for it to work well, so I am going to have to make the assumption that the camera on the Q cell phone has a bit of a wide angle lens.  I suspect that such a lens would be very undesirable on the Aimtrak, due to the distortion inherent with inexpensive wide angle lenses, so it probably has a much narrower field of view, which provides a more uniform image.

If this isn't the case, however, I'm having a hard time understanding why a second set of emitters, with a reversed pattern (i.e. same board, flipped) or not, wouldn't allow one to get quite a bit closer than is currently possible.  

RandyT

You said that you can get all four corners on the screen at a distance of 27in.
Now imagine a line drawn from the phones center axis, to the screen (or even fit a laser pointer to the phone). Move the phone until the pointer reaches the edge of the screen. You will not be able to see all four corners on the camera picture.
Now move back until you are able to get the pointer to hit all four corners while still being able to see the opposite corner on the picture...
Actually you will need to move back twice the distance.

If the sensor can work using one LED, at least for part of its field, then the disctance does not have to be doubled. Adding another board below the screen might increase the vertical field of view if the sensor could use either the top or botom board, but would not affect the horizontal field of view. Its true that the lower sensor might provide more information, at least when the camera can see both boards. But its not practical because at certain locations of the pointer, the camera will only be able to see one LED (ie one of the four LED groups) and would not know which one its looking at unless there is active control of each LED as in the Sega system.

[brandon]

I took down my earlier post, because I assumed after the fact that the CMOS sensor was unable to ever see two emitter boards at once, unless it was a good distance from the screen.  When aimed at the center of the screen, the CMOS sensor would absolutely need to have more than one set of emitters in frame for any of the advanced stuff related to multiple emitters to be done.

I think there's not enough information about the hardware available, specifically the field-of-view of the CMOS sensor / lens, to know whether this system can be improved at all.  Andy mentioned something about a cell phone camera earlier, so I did some testing with my (t)rusty old Motorola Q.  One thing that is very interesting is that I can get the entire monitor in frame, with some extra space above and below for emitters from about 27" in front of the emitters.  This is about 2' closer than where I need to use the Aimtrak for it to work well, so I am going to have to make the assumption that the camera on the Q cell phone has a bit of a wide angle lens.  I suspect that such a lens would be very undesirable on the Aimtrak, due to the distortion inherent with inexpensive wide angle lenses, so it probably has a much narrower field of view, which provides a more uniform image.

If this isn't the case, however, I'm having a hard time understanding why a second set of emitters, with a reversed pattern (i.e. same board, flipped) or not, wouldn't allow one to get quite a bit closer than is currently possible. 

RandyT

You said that you can get all four corners on the screen at a distance of 27in.
Now imagine a line drawn from the phones center axis, to the screen (or even fit a laser pointer to the phone). Move the phone until the pointer reaches the edge of the screen. You will not be able to see all four corners on the camera picture.
Now move back until you are able to get the pointer to hit all four corners while still being able to see the opposite corner on the picture...
Actually you will need to move back twice the distance.

If the sensor can work using one LED, at least for part of its field, then the disctance does not have to be doubled. Adding another board below the screen might increase the vertical field of view if the sensor could use either the top or botom board, but would not affect the horizontal field of view. Its true that the lower sensor might provide more information, at least when the camera can see both boards. But its not practical because at certain locations of the pointer, the camera will only be able to see one LED (ie one of the four LED groups) and would not know which one its looking at unless there is active control of each LED as in the Sega system.

I wonder if the Sega system has all LEDs "steady on" but strobes them when its unsure which LED its looking at... Because if the sensor can only refresh at, say 60hz then each corner would only be updated every 15hz and that would cause noticeable lag.  I know that's not really helpful in regard to the AimTrak but I'm curious

[RandyT]

You said that you can get all four corners on the screen at a distance of 27in.
Now imagine a line drawn from the phones center axis, to the screen (or even fit a laser pointer to the phone). Move the phone until the pointer reaches the edge of the screen. You will not be able to see all four corners on the camera picture.

Obviously. but I can still see the center area where the emitters are positioned, which is really the only thing that's important. 

But its not practical because at certain locations of the pointer, the camera will only be able to see one LED (ie one of the four LED groups) and would not know which one its looking at unless there is active control of each LED as in the Sega system.

I'm not suggesting the more complex implementation as others are.  There is middle ground that I'm not sure you are seeing, so I'll give you a scenario and a question;

Given the notion that the center screen position of the camera shows an emitter at very near the top of the image and one at very near the bottom of the image, at what point could the gun not be able to accurately determine the current emitter if the logic were :

Two Emitters = Don't Care
Above Emitter = Lower Emitter
Below Emitter = Upper Emitter

given the limitation of left to right off-screen reload shots only?

[AndyWarne]

Obviously. but I can still see the center area where the emitters are positioned, which is really the only thing that's important. 

I'm not suggesting the more complex implementation as others are.  There is middle ground that I'm not sure you are seeing, so I'll give you a scenario and a question;

Given the notion that the center screen position of the camera shows an emitter at very near the top of the image and one at very near the bottom of the image, at what point could the gun not be able to accurately determine the current emitter if the logic were :

Two Emitters = Don't Care
Above Emitter = Lower Emitter
Below Emitter = Upper Emitter

given the limitation of left to right off-screen reload shots only?

Dont understand at all. The camera cant always see the center area, it moving with the gun...If its pointing at the top of the screen it loses the bottom emitter and vice versa. The cameran can only see both when the gun is aimed near the center of the screen.

[KissMyWookie]

... But its not practical because at certain locations of the pointer, the camera will only be able to see one LED (ie one of the four LED groups) and would not know which one its looking at unless there is active control of each LED as in the Sega system.

Or if each LED group had a different visible configuration, which is possible with the rotated L-shaped emitter clusters I previously mentioned.

[brandon]

Just ask Johnny Lee  




BTW, What if you had an extra AimTrak facing the user that tracked the GUN?  add an IR LED to the gun that was in a position that could only be seen by the AimTrak facing the users and wouldnt interfere with the sensors in the gun.. 

[KissMyWookie]

BTW, What if you had an extra AimTrak facing the user that tracked the GUN?  add an IR LED to the gun that was in a position that could only be seen by the AimTrak facing the users and wouldnt interfere with the sensors in the gun.. 

That wouldn't work, because it would only see the position of the gun and not where it was pointing.

[brandon]

BTW, What if you had an extra AimTrak facing the user that tracked the GUN?  add an IR LED to the gun that was in a position that could only be seen by the AimTrak facing the users and wouldnt interfere with the sensors in the gun..  

That wouldn't work, because it would only see the position of the gun and not where it was pointing.

I said in addition to the one already IN the gun.  Then it know where the gun was pointing and where from.  Thus you could move or hand the gun to your short little brother.

[KissMyWookie]

BTW, What if you had an extra AimTrak facing the user that tracked the GUN?  add an IR LED to the gun that was in a position that could only be seen by the AimTrak facing the users and wouldnt interfere with the sensors in the gun..  

That wouldn't work, because it would only see the position of the gun and not where it was pointing.

I said in addition to the one already IN the gun.  Then it know where the gun was pointing and where from.  Thus you could move or hand the gun to your short little brother.

Okay, I see what you mean - but I see another problem in that approach...
The CCD would not be able to distinguish between depth and height, as holding the gun lower would look the same to the sensor as being closer to the screen (this is assuming the CCD is absolutely perpendicular to your floor and not tilted even 1degree).

[brandon]

You guys may find this interesting..  

http://muonics.net/blog/index.php?postid=26




a webcam tracking a laser pointer.

[RandyT]

Dont understand at all. The camera cant always see the center area, it moving with the gun...If its pointing at the top of the screen it loses the bottom emitter and vice versa. The cameran can only see both when the gun is aimed near the center of the screen.

Correct.  Right now, the system works pretty well with only one emitter in view.  The only time it would ever see two is of the camera were pointed somewhere in the middle of the screen.

The reason one needs to be so far back is because a single top mounted emitter gets lost, at close distances, as soon as one starts moving downward.  This wouldn't matter with a dual emitter system, because the gun would have switched over to the bottom emitter by time the upper one was lost off image, and vice versa.

With some calibration tricks, this arrangement would not only let one get closer, but should be more accurate, as each half of the screen has a reference which is more local to the actual position of the gun.

RandyT

[AndyWarne]

Dont understand at all. The camera cant always see the center area, it moving with the gun...If its pointing at the top of the screen it loses the bottom emitter and vice versa. The cameran can only see both when the gun is aimed near the center of the screen.

Correct.  Right now, the system works pretty well with only one emitter in view.  The only time it would ever see two is of the camera were pointed somewhere in the middle of the screen.

The reason one needs to be so far back is because a single top mounted emitter gets lost, at close distances, as soon as one starts moving downward.  This wouldn't matter with a dual emitter system, because the gun would have switched over to the bottom emitter by time the upper one was lost off image, and vice versa.

With some calibration tricks, this arrangement would not only let one get closer, but should be more accurate, as each half of the screen has a reference which is more local to the actual position of the gun.

RandyT

The sensor is angled upwards so that there is approx the same amount of leeway top and bottom.
If the gun needs to work with either a top or bottom LED bar, it needs to know which one its looking at. You mentioned reversing the lower one. Determining which bar is which, using the reversal, means that both the left and right hand LED group of each bar needs to be visible at all times. This means the horizontal field of view would be worse than at present because in the current design only one LED needs to be visible and this one LED is located approx in the center of the overall field of view. The other led is used only when it can be seen, and is not used for the main X-Y tracking but only for correction purposes.

[Endaar]

I have to say all this back and forth on the technical workings of a system like this and the considerations that go into a product's design are fascinating.

[bkenobi]

BTW, What if you had an extra AimTrak facing the user that tracked the GUN?  add an IR LED to the gun that was in a position that could only be seen by the AimTrak facing the users and wouldnt interfere with the sensors in the gun.. 

If I understand your suggestion, there is already a gun that does this.  I don't think it was received well though.  I can't remember what it was called, but it basically had a web cab sitting on the monitor and the gun had a big growth on the top.  I think there was an LED in the tip of the gun and another in the tumor.  Supposedly, it could work with multiple guns.  I just wish I could find a link.

[brandon]

BTW, What if you had an extra AimTrak facing the user that tracked the GUN?  add an IR LED to the gun that was in a position that could only be seen by the AimTrak facing the users and wouldnt interfere with the sensors in the gun.. 

If I understand your suggestion, there is already a gun that does this.  I don't think it was received well though.  I can't remember what it was called, but it basically had a web cab sitting on the monitor and the gun had a big growth on the top.  I think there was an LED in the tip of the gun and another in the tumor.  Supposedly, it could work with multiple guns.  I just wish I could find a link.

just do it like this

[Turnarcades]

Just ask Johnny Lee

Saw this guy's first Wiimote videos on youtube time ago. I really wish he'd extend this concept and get a game design around that pseudo-3D idea.

[extendedplayarcade]

I am just posting so I can subscribe to this thread. 

[Neverending Project]

So last night I cam across a game called One Shot One Kill. Anyone ever try it or hear of it? It's a lightgun game, and is not particularly difficult. But it's the content that I found amusing. Ginsu - where are you on your alphabetical tour?

The challenges are things like shooting the people who jump the turnstiles in the subway, but not the people who pay. Or shooting kids as they run away from home, so they will turn around and go back. Or on a subway shooting the hands of creeps who are reaching toward a young girl in a miniskirt and schoolgirl outfit.

I just found the whole thing to be amusing.

[Ginsu Victim]

So last night I cam across a game called One Shot One Kill. Anyone ever try it or hear of it? It's a lightgun game, and is not particularly difficult. But it's the content that I found amusing. Ginsu - where are you on your alphabetical tour?

Haven't played anything in a while since my gun is out of order, so I'm still in the N's (Ninja Baseball Batman!).

The challenges are things like shooting the people who jump the turnstiles in the subway, but not the people who pay. Or shooting kids as they run away from home, so they will turn around and go back. Or on a subway shooting the hands of creeps who are reaching toward a young girl in a miniskirt and schoolgirl outfit.

I just found the whole thing to be amusing.

Sounds awesome. I've gotta fix my gun and try it.

[thatitalian]

 

[Silas (son of Silas)]

+ 1

[AndyWarne]

Speaking of PC applications, any word on a new version of the AimTrak software?

Just an update: Still working on the new version and also new firmware. There are quite a few changes so its taking a while. Will be ready as soon as possible.

Andy

[Ginsu Victim]

Thanks for the update

[jlfreund]

Although curvature of the CRT may be responsible for some of the inaccuracy at the left/right of the monitor, it seems like the accuracy is best inside a "V" from the center to the top two corners of the screen.  Ie, nearest the top sensor bar.

I'm wondering if it should be possible to duplicate that accuracy in the other quadrants just by using 4 sensor bars?  It seems like it should be easy from an algorithmic point of view to just report the location from the nearest sensor bar and ignore the rest of the sensors.

Since the sensor bar hardware is relatively cheap, I would expect the improvement would be worthwhile with respect to a $200 dual gun setup.