update 4.5.18 about sltk

[jhdewitt]

Found a c270 and will take a shot at getting it working with uvc mjpg-streamer module on an rpi this weekend. 720p images should process significantly faster on arm box than 5-8mp ones!

[prahjister]

Awesome. I ended up getting a c920 off Amazon. I have been using brown University structured light with some success.

[jhdewitt]

Ah, yay the c920 is the other usb cam on hand! Pretty sharp imo. Want reproducible results. Glad to hear of success :)

[prahjister]

I didn't want to pressure you. But I am still standing by.

I wish this one took your approach to calibration. Seems better

[jhdewitt]

Ok it seems like it's working now, was a simple mistake. This update works on my end to make -timestamp option function with input_uvc.so: jhdewitt/mjpg-streamer@de757ce

mjpg_streamer -i "input_uvc.so -f 30 -r HD -timestamp" -o "output_http.so"

with c920 I get:
1523665657.559277 : frame# 1730 (0.205 late) (0.204 dt = 4fps) ( 47587 B)

also tried with ps3 eye and it had very low frame delay
1523665143.702563 : frame# 0736 (0.034 late) (0.030 dt = 33fps) ( 18912 B)

[jhdewitt]

now back to fixing all the other broken things ;)

[prahjister]

Fantastic news. I will get to testing here.

What all is broken?

[prahjister]

I know this is off topic but after messing with alignment in mesh lab. Your ply tools look very interesting. But how do you set the noise threshold and does it ignore the surface the object is on?

[jhdewitt]

The current public repo should function fine, but I have been working on refactoring stuff to get the code into a more sane and less rat nest like state. This will allow me to add features and stay sane while doing so.

Re: ply tools, the current idea is that the scanner would be moved infrequently in relation to the static background. My current way of cleaning up scans is as follows:

• use plymerge to overlap all scans in a rotation
• manually delete object and save as PLY "background reference"

or

• make PLY of scene with no object in place and save as "background reference"

then,

• use plytrim on each scan in a rotation, specifying "background reference" file (any point in scan that has a close neighbor in reference file will be omitted)

• take resulting trimmed files (only object should remain) and feed into plyalign to get rough fused pointcloud

• import combined pointcloud into meshlab or cloudcompare and apply poisson filter to get manifold mesh

this is the general procedure i followed for the shell example in the readme.

[jhdewitt]

sorry for long winded explanation

noise is determined by a manually created "background reference". basically run once with no object then save until hardware is moved.

[jhdewitt]

ps the autofocus on the c920 is on by default and interferes with the image capture process. trying to figure out how to disable it.

This thread was the closest I could find on short notice.
https://sourceforge.net/p/mjpg-streamer/discussion/739917/thread/5f8681a5/

This function is probably accessed through dyncontrol but how is totally unclear to me https://github.com/jacksonliam/mjpg-streamer/blob/master/mjpg-streamer-experimental/plugins/input_uvc/dynctrl.h