[libcamera-devel] [PATCH v6 06/10] ipa: rkisp1: Add AF algorithm based on AfHillClimbing
Laurent Pinchart
laurent.pinchart at ideasonboard.com
Wed May 31 19:53:37 CEST 2023
Hi Daniel,
On Fri, May 26, 2023 at 11:19:02AM +0200, Daniel Semkowicz wrote:
> On Wed, Apr 26, 2023 at 6:04 AM Laurent Pinchart wrote:
> > On Wed, Apr 26, 2023 at 06:46:58AM +0300, Laurent Pinchart via libcamera-devel wrote:
> > > On Fri, Mar 31, 2023 at 10:19:26AM +0200, Daniel Semkowicz via libcamera-devel wrote:
> > > > Add contrast based auto focus implementation for rkisp1 platform using
> > > > the common AfHillClimbing algorithm.
> > > >
> > > > Rockchip ISP AF block allows calculation of sharpness and luminance
> > > > in up to three user defined windows. If no windows are set, there are
> > > > some default settings applied for the first window and exposed through
> > > > the driver. For each frame, use the sharpness value calculated for this
> > > > default window and feed the hill climbing algorithm with them. Then set
> > > > the lens position to value calculated by the algorithm.
> > > >
> > > > Signed-off-by: Daniel Semkowicz <dse at thaumatec.com>
> > > > ---
> > > > src/ipa/rkisp1/algorithms/af.cpp | 121 ++++++++++++++++++++++++++
> > > > src/ipa/rkisp1/algorithms/af.h | 43 +++++++++
> > > > src/ipa/rkisp1/algorithms/meson.build | 1 +
> > > > src/ipa/rkisp1/ipa_context.cpp | 13 +++
> > > > src/ipa/rkisp1/ipa_context.h | 5 ++
> > > > 5 files changed, 183 insertions(+)
> > > > create mode 100644 src/ipa/rkisp1/algorithms/af.cpp
> > > > create mode 100644 src/ipa/rkisp1/algorithms/af.h
> > > >
> > > > diff --git a/src/ipa/rkisp1/algorithms/af.cpp b/src/ipa/rkisp1/algorithms/af.cpp
> > > > new file mode 100644
> > > > index 00000000..fde924d4
> > > > --- /dev/null
> > > > +++ b/src/ipa/rkisp1/algorithms/af.cpp
> > > > @@ -0,0 +1,121 @@
> > > > +/* SPDX-License-Identifier: LGPL-2.1-or-later */
> > > > +/*
> > > > + * Copyright (C) 2023, Theobroma Systems
> > > > + *
> > > > + * af.cpp - RkISP1 AF hill climbing based control algorithm
> > >
> > > s/climbing based/climbing-based/
> > >
> > > > + */
> > > > +
> > > > +#include "af.h"
> > > > +
> > > > +/**
> > > > + * \file af.h
> > >
> > > Missing \brief
> > >
> > > > + */
> > > > +
> > > > +namespace libcamera {
> > > > +
> > > > +namespace ipa::rkisp1::algorithms {
> > > > +
> > > > +/**
> > > > + * \class Af
> > > > + * \brief AF contrast based hill climbing control algorithm for RkISP platforms
> > >
> > > s/contrast based/contrast-based/
> > >
> > > > + *
> > > > + * Auto focus algorithm for RkISP platforms, based on the common hill climbing
> > > > + * auto focus implementation (libcamera::ipa::algorithms::AfHillClimbing).
> > > > + *
> > > > + * This is the platform layer of the algorithm.
> > > > + *
> > > > + * Tuning file parameters:
> > > > + * - **wait-frames-lens:** Number of frames that should be skipped when lens
> > >
> > > rkisp1 tuning files use camelCase for parameter names.
>
> I will correct it.
>
> > > > + * position is changed. Lens movement takes some time and statistics measured
> > > > + * during the lens movement are unstable. Currently there is no way to know
> > > > + * when lens movement finished and this is a workaround for this. Wait a fixed
> > > > + * amount of time on each movement. This parameter should be set according
> > > > + * to the worst case - the number of frames it takes to move lens between
> > > > + * limit positions.
> > > > + * .
> > >
> > > Stray period ?
>
> The same as in the 04/10, dot is used to explicitly mark the end of
> the list in the Doxygen.
>
> > > > + * \sa libcamera::ipa::algorithms::AfHillClimbing for additional tuning
> > > > + * parameters.
> > > > + *
> > > > + * \todo Model the lens delay as number of frames required for the lens position
> > > > + * to stabilize in the CameraLens class.
> > >
> > > More than that, we should take the lens movement into account in the
> > > algorithm to avoid skipping frames at all.
>
> Yes, I agree. We should only base on the information if the lens
> movement has finished instead of skipping frames in the AF. I will
> improve that note.
>
> > > > + */
> > > > +
> > > > +LOG_DEFINE_CATEGORY(RkISP1Af)
> > > > +
> > > > +/**
> > > > + * \copydoc libcamera::ipa::Algorithm::init
> > > > + */
> > > > +int Af::init(IPAContext &context, const YamlObject &tuningData)
> > > > +{
> > > > + const auto &lensConfig = context.configuration.lens;
> > > > + if (!lensConfig) {
> > > > + LOG(RkISP1Af, Error) << "Lens not found";
> > > > + return 1;
> > > > + }
> > > > +
> > > > + waitFramesLens_ = tuningData["wait-frames-lens"].get<uint32_t>(1);
> > > > +
> > > > + LOG(RkISP1Af, Debug) << "waitFramesLens_: " << waitFramesLens_;
> > > > +
> > > > + return af.init(lensConfig->minFocusPosition,
> > > > + lensConfig->maxFocusPosition, tuningData);
> > > > +}
> > > > +
> > > > +/**
> > > > + * \copydoc libcamera::ipa::Algorithm::configure
> > > > + */
> > > > +int Af::configure(IPAContext &context,
> > > > + const IPACameraSensorInfo &configInfo)
> > > > +{
> > > > + /*
> > > > + * Lens position is unknown at the startup, so initialize
> > > > + * the current position to something out of range.
> > > > + */
> > > > + context.activeState.af.lensPosition =
> > > > + context.configuration.lens->maxFocusPosition + 1;
> > > > +
> > > > + af.configure(configInfo.outputSize);
> > > > + return 0;
> > > > +}
> > > > +
> > > > +/**
> > > > + * \copydoc libcamera::ipa::Algorithm::queueRequest
> > > > + */
> > > > +void Af::queueRequest([[maybe_unused]] IPAContext &context,
> > > > + [[maybe_unused]] const uint32_t frame,
> > > > + [[maybe_unused]] IPAFrameContext &frameContext,
> > > > + const ControlList &controls)
> > > > +{
> > > > + af.queueRequest(controls);
> > > > +}
> > > > +
> > > > +/**
> > > > + * \copydoc libcamera::ipa::Algorithm::process
> > > > + */
> > > > +void Af::process(IPAContext &context, [[maybe_unused]] const uint32_t frame,
> > > > + [[maybe_unused]] IPAFrameContext &frameContext,
> > > > + const rkisp1_stat_buffer *stats,
> > > > + [[maybe_unused]] ControlList &metadata)
> > > > +{
> > > > + uint32_t sharpness = stats->params.af.window[0].sum;
> > > > + uint32_t luminance = stats->params.af.window[0].lum;
> > > > +
> > > > + LOG(RkISP1Af, Debug)
> > > > + << "lensPosition: " << context.activeState.af.lensPosition
> > > > + << ", Sharpness: " << sharpness
> > >
> > > Maybe contrast, as everything in this series talks about contrast-based
> > > AF ?
>
> Log category for this debug message is "RkISP1Af". Isn't it easier
> for debugging, if We stick with the rkisp naming in the platform related
> code?
I was talking about "Sharpness", not the category :-) But the ISP
documentation talks about sharpness, so I'm fine with that term.
> > > > + << ", Luminance: " << luminance;
> > >
> > > The luminance isn't used, so you could drop it.
> >
> > Or better, you should use it :-) The luminance is meant to normalize the
> > sharpness values.
>
> Can We introduce the normalization later as an improvement?
I've explained in the review of the next patch how to handle the shift
calculation. The normalization should be equally simple (in terms of
code at least).
The idea is that the sharpness is proportional to the square of the
pixel values, so higher illumination will result in higher sharpness. To
compensate for the illumination, you can simply divide the sharpness by
the square of the mean illumination over the window:
- Multiply the illumination computed by the hardware by the shift factor
to make it independent of the window size.
- Divide the resulting value by the window size in pixels, to obtain a
mean illumination in the [0.0, 1.0] range.
- Divide the sharpness by the square of the mean illumination. Make sure
the mean illumination value is > 0 to avoid divisions by 0.
It would be nice if you could do so already, as it should just be a few
lines of code.
> > > > +
> > > > + int32_t newLensPosition = af.process(sharpness);
> > > > +
> > > > + if (newLensPosition != context.activeState.af.lensPosition) {
> > > > + context.activeState.af.lensPosition = newLensPosition;
> > > > + context.activeState.af.applyLensCtrls = true;
> > > > + af.skipFrames(waitFramesLens_);
> > >
> > > I'm tempted to implement frame skipping in this class instead of the
> > > ipa::algorithms::AfHillClimbing class. It would be simpler, we could
> > > just skip calling process(). On the other hand, I suppose
> > > ipa::algorithms::AfHillClimbing will need to change when we'll improve
> > > the mechanism that takes lens movement into account, and it's hard to
> > > predict how the interface will change.
>
> So maybe it would be simpler to use the current approach and improve
> it later when needed, when the interface become clear.
OK. Let's keep frame skipping where it is, but I'd like frame skipping
to cover the lens movement only if possible, not the other delays. That
may be lots of yak shaving though, so this could be done on top too.
Would you agree to give it a try once this series gets merged ? :-)
> > > > + }
> > > > +}
> > >
> > > The IPA module should also report the lens position in metadata.
> > >
> > > > +
> > > > +REGISTER_IPA_ALGORITHM(Af, "Af")
> > > > +
> > > > +} /* namespace ipa::rkisp1::algorithms */
> > > > +
> > > > +} /* namespace libcamera */
> > > > diff --git a/src/ipa/rkisp1/algorithms/af.h b/src/ipa/rkisp1/algorithms/af.h
> > > > new file mode 100644
> > > > index 00000000..3ba66d38
> > > > --- /dev/null
> > > > +++ b/src/ipa/rkisp1/algorithms/af.h
> > > > @@ -0,0 +1,43 @@
> > > > +/* SPDX-License-Identifier: LGPL-2.1-or-later */
> > > > +/*
> > > > + * Copyright (C) 2023, Theobroma Systems
> > > > + *
> > > > + * af.h - RkISP1 AF hill climbing based control algorithm
> > >
> > > s/climbing based/climbing-based/
> > >
> > > > + */
> > > > +
> > > > +#pragma once
> > > > +
> > > > +#include <linux/rkisp1-config.h>
> > > > +
> > > > +#include "libipa/algorithms/af_hill_climbing.h"
> > > > +
> > > > +#include "algorithm.h"
> > > > +
> > > > +namespace libcamera {
> > > > +
> > > > +namespace ipa::rkisp1::algorithms {
> > > > +
> > > > +class Af : public Algorithm
> > > > +{
> > > > +public:
> > > > + int init(IPAContext &context, const YamlObject &tuningData) override;
> > > > + int configure(IPAContext &context,
> > > > + const IPACameraSensorInfo &configInfo) override;
> > > > + void queueRequest(IPAContext &context, uint32_t frame,
> > > > + IPAFrameContext &frameContext,
> > > > + const ControlList &controls) override;
> > > > + void process(IPAContext &context, uint32_t frame,
> > > > + IPAFrameContext &frameContext,
> > > > + const rkisp1_stat_buffer *stats,
> > > > + ControlList &metadata) override;
> > > > +
> > > > +private:
> > > > + ipa::algorithms::AfHillClimbing af;
> > >
> > > s/af/af_/
> > >
> > > > +
> > > > + /* Wait number of frames after changing lens position */
> > > > + uint32_t waitFramesLens_ = 0;
> > > > +};
> > > > +
> > > > +} /* namespace ipa::rkisp1::algorithms */
> > > > +
> > > > +} /* namespace libcamera */
> > > > diff --git a/src/ipa/rkisp1/algorithms/meson.build b/src/ipa/rkisp1/algorithms/meson.build
> > > > index 93a48329..ab7e44f3 100644
> > > > --- a/src/ipa/rkisp1/algorithms/meson.build
> > > > +++ b/src/ipa/rkisp1/algorithms/meson.build
> > > > @@ -1,6 +1,7 @@
> > > > # SPDX-License-Identifier: CC0-1.0
> > > >
> > > > rkisp1_ipa_algorithms = files([
> > > > + 'af.cpp',
> > > > 'agc.cpp',
> > > > 'awb.cpp',
> > > > 'blc.cpp',
> > > > diff --git a/src/ipa/rkisp1/ipa_context.cpp b/src/ipa/rkisp1/ipa_context.cpp
> > > > index aea99299..d80a7b1b 100644
> > > > --- a/src/ipa/rkisp1/ipa_context.cpp
> > > > +++ b/src/ipa/rkisp1/ipa_context.cpp
> > > > @@ -139,6 +139,19 @@ namespace libcamera::ipa::rkisp1 {
> > > > * member may be read by any algorithm, but shall only be written by its owner.
> > > > */
> > > >
> > > > +/**
> > > > + * \var IPAActiveState::af
> > > > + * \brief State for the Automatic Focus Control algorithm
> > > > + *
> > > > + * \var IPAActiveState::af.lensPosition
> > > > + * \brief Lens position calculated by the AF algorithm
> > > > + *
> > > > + * \var IPAActiveState::af.applyLensCtrls
> > > > + * \brief Whether the lens position should be applied
> > > > + *
> > > > + * If true, IPA should send new controls to the PH to set new lens position.
> > > > + */
> > > > +
> > > > /**
> > > > * \var IPAActiveState::agc
> > > > * \brief State for the Automatic Gain Control algorithm
> > > > diff --git a/src/ipa/rkisp1/ipa_context.h b/src/ipa/rkisp1/ipa_context.h
> > > > index 65b3fbfe..3dcc5aa0 100644
> > > > --- a/src/ipa/rkisp1/ipa_context.h
> > > > +++ b/src/ipa/rkisp1/ipa_context.h
> > > > @@ -62,6 +62,11 @@ struct IPASessionConfiguration {
> > > > };
> > > >
> > > > struct IPAActiveState {
> > > > + struct {
> > > > + int32_t lensPosition;
> > > > + bool applyLensCtrls;
> > >
> > > I suppose applyLens will be used later in the series.
>
> Should I move it to the later patches?
No, it's OK, it was a note for myself.
> > >
> > > > + } af;
> > > > +
> > > > struct {
> > > > struct {
> > > > uint32_t exposure;
--
Regards,
Laurent Pinchart
More information about the libcamera-devel
mailing list