Previous Thread
Next Thread
Print Thread
Rate This Thread
Hop To
Page 4 of 4 1 2 3 4
#2726147 - 05/17/09 12:15 AM Re: TrackIR 5 ***** [Re: Dslyecxi]  
Joined: Dec 2002
Posts: 1,733
SpyDoc Offline
Member
SpyDoc  Offline
Member

Joined: Dec 2002
Posts: 1,733
Central USA
Well, maybe a 2.66 quad (not I7) is not exactly mid grade..., or is it?

biggrin

Probably depends on who/what you are used to or comparing to.

Nice explanation--NP Seth.

smile


SpyDoc out cool

My new DigitalStormOnline system:
i5-2500K--8GB Ram--AMD Radeon 6950
Win7--MOMO--MS FFB2--TIR5--27.5" Hanns-G
Inline advert (2nd and 3rd post)

#2729209 - 05/21/09 05:17 PM Re: TrackIR 5 [Re: NaturalPoint_Seth]  
Joined: Mar 2006
Posts: 2,454
MajorMagee Offline
Member
MajorMagee  Offline
Member

Joined: Mar 2006
Posts: 2,454
Dayton, OH
Originally Posted By: NaturalPoint_Seth
Originally Posted By: D13-th_Korn
...

TrackIR 5 does have the raw and reporting resolution we claim—otherwise we would not claim it. We respect our customers and are not misleading them with our marketing. Our measurements are based on engineering and physical testing.

Reporting resolution is not the same as raw sensor resolution, so we’re not implying that TrackIR 5 is a gigapixel camera. Neither are our numbers the result of software interpolation. In fact, TrackIR is not even a “camera” in the way you are describing, and its function is different than standard camera imaging. We’re not out to capture what an image looks like. We’re looking for the location of marker centroids.

When we describe sub-pixel resolution, we are defining how accurately we can discern the center of the marker in terms of pixels. It is a unit of measurement. Although our raw sensor size is 640 x 480, TrackIR 5 can actually measure movement that is much smaller than a pixel—as small as 1/150th of a pixel. Spread that across 640 pixels and what do you get? A range of 96,000 detectable object locations.

That’s not the same as saying that we have a 6 gigapixel camera. If you have a ruler with 12 inches on it, and you measure an object that falls in between the inch marks, accurate measurement requires sub-inch units. The principle is the same with our grayscale image processing. The center of a marker usually lies between two adjacent pixel locations. We use sub-pixel analysis to accurately find that center down to 1/150th of a pixel's accuracy. Clearly we do not transmit the entire image. The entire benefit of having a TrackIR is that we make a custom, purpose-built product for tracking markers. It processes the image on-board at 120 FPS, which is a very large amount of data, and then only transmits the relevant information to the PC. Thus, there is very little load on the PC.

Consumer digital cameras used for photography sometimes use sub-pixel resolution when creating a viewable image, in which case adjacent pixels may be interpolated. That is not what we are doing—we have a grayscale imager, and we don’t use a Bayer pattern. Rather, we are describing how our new processing has advanced the effective resolution of finding a marker and reporting its center to the software. There are many ways to do this, and our method has achieved the resolution we report.

If you don't believe our figures, please feel free to test them yourself with our free OptiTrack SDK and a micrometer.


The better way to explain this is that the sensor is taking the average of a collection of pixels, and that allows for the interpolation of sub-pixel locations. TrackIR4 averaged 20 pixel locations, and TrackIR5 averages 150.

So TrackIR4 was 355 x 20 x 288 x 20 = 7,100 x 5,760 and TrackIR5 is 640 x 150 x 480 x 150 = 96,000 x 72,000 discreet locations that could be calculated.

The problem is that if the sensor is not seeing the full 150 active pixels then the quality of the average falls off. Theoreticaly the absolute minimum number of active pixels to get 6 degrees of freedom is just 3. I would imagine the software has a much higher minimum cut-off than that, but you would only get 640 x 3 x 480 x 3 = 1920 x 1440

Here's a simple example. Let's say we have a 10 x 10 grid of sensors, and we want to know the center of three points within the grid.
Point 1 = 2 , 6
Point 2 = 1 , 9
Point 3 = 7 , 2
Average = 10/3 , 17/3 = 3 1/3 , 5 2/3

so our integer x , y locations have each been divided into fractional thirds. Rather than just 100 physical locations we can now calculate 900 average locations of 3 signals.


Service To The Line,
On The Line,
On Time

US Army Ordnance Corps.
#2731819 - 05/26/09 04:17 PM Re: TrackIR 5 [Re: MajorMagee]  
Joined: Jan 2009
Posts: 22
NaturalPoint_Seth Offline
NP Marketing Manager
NaturalPoint_Seth  Offline
NP Marketing Manager
Junior Member

Joined: Jan 2009
Posts: 22
Originally Posted By: MajorMagee

The better way to explain this is that the sensor is taking the average of a collection of pixels, and that allows for the interpolation of sub-pixel locations. TrackIR4 averaged 20 pixel locations, and TrackIR5 averages 150.

So TrackIR4 was 355 x 20 x 288 x 20 = 7,100 x 5,760 and TrackIR5 is 640 x 150 x 480 x 150 = 96,000 x 72,000 discreet locations that could be calculated.

The problem is that if the sensor is not seeing the full 150 active pixels then the quality of the average falls off. Theoreticaly the absolute minimum number of active pixels to get 6 degrees of freedom is just 3. I would imagine the software has a much higher minimum cut-off than that, but you would only get 640 x 3 x 480 x 3 = 1920 x 1440

Here's a simple example. Let's say we have a 10 x 10 grid of sensors, and we want to know the center of three points within the grid.
Point 1 = 2 , 6
Point 2 = 1 , 9
Point 3 = 7 , 2
Average = 10/3 , 17/3 = 3 1/3 , 5 2/3

so our integer x , y locations have each been divided into fractional thirds. Rather than just 100 physical locations we can now calculate 900 average locations of 3 signals.

Actually, that isn't quite how it works. We aren't averaging 20 pixel locations to get 1/20th pixel accuracy. Instead, we use sophisticated weighting of all the pixels related to a detected marker. While there is a relation between the number of pixels in a marker and the accuracy with which the marker's position can be calculated, it isn't a 1:1 correlation. Nor is the correlation linear.


NP Marketing Manager
seth @ naturalpoint . com
Page 4 of 4 1 2 3 4

Moderated by  RacerGT 

Quick Search
Recent Articles
Support SimHQ

If you shop on Amazon use this Amazon link to support SimHQ
.
Social


Recent Topics
Carnival Cruise Ship Fire....... Again
by F4UDash4. 03/26/24 05:58 PM
Baltimore Bridge Collapse
by F4UDash4. 03/26/24 05:51 PM
The Oldest WWII Veterans
by F4UDash4. 03/24/24 09:21 PM
They got fired after this.
by Wigean. 03/20/24 08:19 PM
Grown ups joke time
by NoFlyBoy. 03/18/24 10:34 PM
Anyone Heard from Nimits?
by F4UDash4. 03/18/24 10:01 PM
RIP Gemini/Apollo astronaut Tom Stafford
by semmern. 03/18/24 02:14 PM
10 years after 3/8/2014
by NoFlyBoy. 03/17/24 10:25 AM
Copyright 1997-2016, SimHQ Inc. All Rights Reserved.

Powered by UBB.threads™ PHP Forum Software 7.6.0