| Age | Commit message (Collapse) | Author |
|---|
|
It looks weird otherwise.
|
|
|
|
|
|
This fixes forced hidpi
|
|
The check is becoming more expensive with more positions checked.
|
|
We can't depend on M_PI existing after including cmath.
|
|
|
|
Sadly, it's only implemented right now on win32.
Remove "set enabled" code for the video widget since it only works for
explicit window minimization, not covering by other windows.
|
|
|
|
|
|
The height was zero on the test video. The ROI check only saw if width
is at least zero. Check for both to be greater than one.
Video provided by: @kblomster
Issue: #375
Also, fix minor issues:
- nix vars that can be const static in function scope
- don't call solvepnp twice where obj_points shift will do
- don't do bounds checking on vector elt access
- respect sprintf varargs type size; change to snprintf for no reason
- fix clamp-to-image logic
- set proper alpha for fps estimation
|
|
Some new matrix element type requirements came up after opencv update
Also,
- switch to matrices of known sizes wherever possible
- split into functions for readability
- use member variables rather than locals to ensure heap allocation
There was also breakage wrt "unknown element type" deep in opencv that
only happens on release mode with no symbols. It's unknown to me
whether the issue got fixed or variable reordering made it corrupt
something else. It appears to work however. -fstack-protector-all
doesn't show any errors at all.
@kblomster says it worked in rc49p2. Looks like -fipa-pta has a
miscompilation despite finally working to begin with.
Reported-by: @kblomster
Issue: #375
|
|
Less overhead this way.
|
|
|
|
- Remove "this->" where it's not needed. Possibly rename shadowed vars.
- Don't reload the options bundle manually since `options::opts' exists
for that very reason.
- Remove '^ \+$' whitespace
- :retab
|
|
Adjust usages.
|
|
It's automatic now.
|
|
In particular PT's calibration interval was 10 Hz which is too much by
far. Make both 4 Hz.
Issue: #344
|
|
libaruco doesn't clamp the min/max sizes but throws an exception.
Reported-by: @duamutefmc
Data-by: @Emton
Issue: #329
Closes #329
|
|
|
|
|
|
When requesting specific modules, opencv doesn't need the others built.
|
|
|
|
Really fast toggling tracking crashed with my PS3 Eye.
|
|
Suitable for 30 Hz camera modes and higher.
Requested-by: @frost555
Issue: #273
|
|
Higher box sizes use more CPU due to the need to convolve a lot.
It looks fine with both high and low exposure on both Logitech C525 and
PS3 Eye webcams.
Issue: #273
|
|
Detection rate stays as good, likely better as before.
@mursey reports in #274 that non-Otsu case eats way more CPU.
|
|
- cycle otsu twice before selecting new box size
don't cycle otsu each frame.
- slower backoff from timeout value before new box size
on occasional detections need at least 5 successes for each failure
to not exceed.
- select more sensible timeout per box size of .35 seconds. this is
enough also with 30 Hz webcams, but too little for 15 Hz modes.
|
|
|
|
|
|
Closes #224
|
