Reply
Thread Tools
Posts: 3,319 | Thanked: 5,610 times | Joined on Aug 2008 @ Finland
#121
Originally Posted by wotevah View Post
Simple physics.
acceleration over time = speed
speed over time = distance
We assume the camera was static at focus time so not hard to calculate how much it moved.
Three challenges: One, the syncing of the accelerometer reads with the exposure - while the exposure itself might take a small fraction of a second, it will be hard to match that up with accel data - leaving you with covering the whole duration of the API call which can be a source for false alarms. Two, rotation. The up-down-left-right style movement can be judged, but you can get a blurry photo with all sensors reading 0 because the movement was angular. Depends on how you hold/press your camera I guess. Three... your subject might be moving in which case the right thing to do is to replicate the motion of the subject AFTER focus.
__________________
Blogging about mobile linux - The Penguin Moves!
Maintainer of PyQt (see introduction and docs), AppWatch, QuickBrownFox, etc
 
Posts: 31 | Thanked: 35 times | Joined on Jun 2010
#122
Originally Posted by geohsia View Post
BTW, do you use dcraw? If you do I want to ask you to try to remove color (chrominance) noise from RAW files with ISO 400+. LR has no affect. it saddens me because the main thing I look for for using RAW.
I used it before but not with the N900. I just tried it now and dcraw generates garbage from the FCamera files though. Something to do with wrong image size somewhere in metadata, dcraw thinks it's 640x480 because of that.

I was going to suggest ufraw or the recently-made-free rawtherapee, looks to be a great program and also uses modified dcraw code with some new demosaicing algorithms. But both crash at this time with the Fcamera files. Going to try again after a new Fcamera release.

Last edited by wotevah; 2010-07-31 at 02:09.
 
Posts: 31 | Thanked: 35 times | Joined on Jun 2010
#123
Originally Posted by attila77 View Post
Three challenges: One, the syncing of the accelerometer reads with the exposure - while the exposure itself might take a small fraction of a second, it will be hard to match that up with accel data - leaving you with covering the whole duration of the API call which can be a source for false alarms. Two, rotation. The up-down-left-right style movement can be judged, but you can get a blurry photo with all sensors reading 0 because the movement was angular. Depends on how you hold/press your camera I guess. Three... your subject might be moving in which case the right thing to do is to replicate the motion of the subject AFTER focus.
I'm sure there are many challenges, I just wanted to throw it out there because it seems plausible. You can have a separate process that reads the accel and keeps track of position offset. The camera app just talks to it and gets some vectors when it needs. Heck, the mapping applications should do that too when the GPS signal is temporarily lost or when you are not moving, to know the last orientation vector and quit bobbing around the spot until you get moving again. But that's another topic.

As far as I know, unlike phase-detect AF which knows which way to move the focus and by how much, contrast-AF can tell neither direction nor amount, so it moves up and down the entire focus scale looking for maximum contrast. A gimmick like this can give contrast-AF a hint of where the new focus point might have moved to from last time good focus was achieved.

Last edited by wotevah; 2010-07-31 at 02:11.
 

The Following User Says Thank You to wotevah For This Useful Post:
Posts: 515 | Thanked: 259 times | Joined on Jan 2010
#124
Originally Posted by wotevah View Post
I was going to suggest ufraw or the recently-made-free rawtherapee, which appears to be a great program (also uses modified dcraw code with some new demosaicing algorithms). But both crash at this time with the Fcamera files. Going to try again after a new Fcamera release.
I'm a nerd but waay too practical to use anything but Lightroom. It's my go to program. I think fCamera really has potential. I spoke with the developer and apparently it is a TIF but it is RAW as well. Not exactly sure what that means but I guess as long as its 16-bit TIFF that's ok, pre-8-bit colorspace conversion.

In Lightroom (and I presume Adobe Camera RAW) you have to use the 2003 development algorithm instead of the 2010 (released with LR 3). I verified it works, but still would rather have 2010 because the 2003 algorithm isn't strong enough to really denoise (chroma) noisy smartphones. The 2010 would do wonders but that isn't working yet. Not sure if its a file metadata thing or an Adobe thing. My guess is the file.

BTW I still like my idea (obviously ;-)) of a software button. The touch sensitivity means you don't have to press hard which means reduced camera movement. It would be nice to have an icon that indicates the camera is in motion so at least we have visual cues. That could definitely use the accelerometer to detect. If they can visually indicate motion only when the motion exceeds that which the shutter speed is able to handle that'd be insane. Different shutter speeds will exhibit blur with different amounts of movement.

The problem with the camera waiting for no motion is that it assumes I never want to pan. There are some shots where I am deliberately in motion or able to maintain motion in a way that is still able to maintain a sharp photo. I'd rather get a shot that's somewhat blurry than no photos at all.
 
Posts: 515 | Thanked: 259 times | Joined on Jan 2010
#125
Originally Posted by wotevah View Post
As far as I know, unlike phase-detect AF which knows which way to move the focus and by how much, contrast-AF can tell neither direction nor amount, so it moves up and down the entire focus scale looking for maximum contrast. A gimmick like this can give contrast-AF a hint of where the new focus point might have moved to from last time good focus was achieved.
This would be quite interesting actually but it doesn't account for subject motion and I think it might actually need a gyroscope for best effect but would still be interesting. Any hints like that even imperfect would probably reduced time needed for AF lock.
 
Posts: 144 | Thanked: 53 times | Joined on Dec 2009
#126
HELP.. for the life of me i can't find the "HDR Camera" of "Lowlight" applications. I have extras-dev enabled.
 
dchky's Avatar
Posts: 549 | Thanked: 299 times | Joined on Jun 2010 @ Australian in the Philippines
#127
Originally Posted by wotevah View Post
Accelerometer-aided focus: if the camera is pointed in the same direction and has focused once then is moved forward or backwards relative to the sensor plane, it should add the movement to the focus distance before focusing again.
Ahh, another person fed up with the inaccuracies present in the N900 accelerometers.

I solved that pesky issue by ripping out the amateurish IC and replaced it with a small 80 pound triple ring laser gyro and an inertial navigation unit. I now have the ability to perform automatic focus recalculations no matter if I move the camera just a few fractions of a millimeter in any direction - or completely around the world several times. Needless to say I have to carry my N900 on a trolley. The internal battery has become somewhat short lived as you might imagine. I know, hard to believe!, so I've coupled it to a small gas powered turbine.
 

The Following User Says Thank You to dchky For This Useful Post:
nwerneck's Avatar
Posts: 304 | Thanked: 233 times | Joined on Jul 2009 @ São Paulo, SP, Brasil
#128
Maybe this has been mentioned, but I couldn't find it yet. When I read the raw data, do I get the actual values from the Bayer pattern? Does that mean we can make an application to perform advanced demosaicing? Is that already in the "to-do" list of fcam apps?...

Info:
http://en.wikipedia.org/wiki/Demosaicing
http://citeseerx.ist.psu.edu/viewdoc...=rep1&type=pdf
 
Posts: 5,795 | Thanked: 3,151 times | Joined on Feb 2007 @ Agoura Hills Calif
#129
Originally Posted by ossipena View Post
yeah, just go home and cry while playing a martyr.
Have you ever noticed that you insult people in about one out of three posts? It's a good thing that you usually know what you're talking about.
__________________
All I want is 40 acres, a mule, and Xterm.
 

The Following 2 Users Say Thank You to geneven For This Useful Post:
Posts: 31 | Thanked: 11 times | Joined on Sep 2009
#130
Originally Posted by ETalvala View Post
We're aware of the problem with dcraw, and it's fixed in SVN - when we release the next version of FCam and FCamera, they'll produce dcraw-compatible DNGs. (it was a one-line fix)
.
Sweet! Any way to fix the existing photos I've taken by chance?
 
Reply

Thread Tools

 
Forum Jump


All times are GMT. The time now is 08:45.