libcamera v0.0.0+3240-f2a18172-dirty (2021-12-20T12:34:46+00:00)
Supporting cameras in Linux since 2019
Classes | Public Member Functions | List of all members
libcamera::ipa::ipu3::algorithms::Awb Class Reference

A Grey world white balance correction algorithm. More...

Inheritance diagram for libcamera::ipa::ipu3::algorithms::Awb:
Inheritance graph
[legend]
Collaboration diagram for libcamera::ipa::ipu3::algorithms::Awb:
Collaboration graph
[legend]

Public Member Functions

int configure (IPAContext &context, const IPAConfigInfo &configInfo) override
 Configure the Algorithm given an IPAConfigInfo. More...
 
void prepare (IPAContext &context, ipu3_uapi_params *params) override
 Fill the params buffer with ISP processing parameters for a frame. More...
 
void process (IPAContext &context, const ipu3_uapi_stats_3a *stats) override
 Process ISP statistics, and run algorithm operations. More...
 

Detailed Description

A Grey world white balance correction algorithm.

The Grey World algorithm assumes that the scene, in average, is neutral grey. Reference: Lam, Edmund & Fung, George. (2008). Automatic White Balancing in Digital Photography. 10.1201/9781420054538.ch10.

The IPU3 generates statistics from the Bayer Down Scaler output into a grid defined in the ipu3_uapi_awb_config_s structure.

In each cell, the ImgU computes for each colour component the average of all unsaturated pixels (below a programmable threshold). It also provides the ratio of saturated pixels in the cell.

The AWB algorithm operates on a coarser grid, made by grouping cells from the hardware grid into zones. The number of zones is fixed to kAwbStatsSizeX x kAwbStatsSizeY. For example, a frame of 1280x720 is divided into 80x45 cells of [16x16] pixels and 16x12 zones of [5x4] cells each (kAwbStatsSizeX=16 and kAwbStatsSizeY=12). If the number of cells isn't an exact multiple of the number of zones, the right-most and bottom-most cells are ignored. The grid configuration is computed by IPAIPU3::calculateBdsGrid().

Before calculating the gains, the algorithm aggregates the cell averages for each zone in generateAwbStats(). Cells that have a too high ratio of saturated pixels are ignored, and only zones that contain enough non-saturated cells are then used by the algorithm.

The Grey World algorithm will then estimate the red and blue gains to apply, and store the results in the metadata. The green gain is always set to 1.

Member Function Documentation

◆ configure()

int libcamera::ipa::ipu3::algorithms::Awb::configure ( IPAContext context,
const IPAConfigInfo &  configInfo 
)
overridevirtual

Configure the Algorithm given an IPAConfigInfo.

Parameters
[in]contextThe shared IPA context
[in]configInfoThe IPA configuration data, received from the pipeline handler

Algorithms may implement a configure operation to pre-calculate parameters prior to commencing streaming.

Configuration state may be stored in the IPASessionConfiguration structure of the IPAContext.

Returns
0 if successful, an error code otherwise

Reimplemented from libcamera::ipa::ipu3::Algorithm.

◆ prepare()

void libcamera::ipa::ipu3::algorithms::Awb::prepare ( IPAContext context,
ipu3_uapi_params *  params 
)
overridevirtual

Fill the params buffer with ISP processing parameters for a frame.

Parameters
[in]contextThe shared IPA context
[out]paramsThe IPU3 specific parameters.

This function is called for every frame when the camera is running before it is processed by the ImgU to prepare the ImgU processing parameters for that frame.

Algorithms shall fill in the parameter structure fields appropriately to configure the ImgU processing blocks that they are responsible for. This includes setting fields and flags that enable those processing blocks.

Reimplemented from libcamera::ipa::ipu3::Algorithm.

◆ process()

void libcamera::ipa::ipu3::algorithms::Awb::process ( IPAContext context,
const ipu3_uapi_stats_3a *  stats 
)
overridevirtual

Process ISP statistics, and run algorithm operations.

Parameters
[in]contextThe shared IPA context
[in]statsThe IPU3 statistics and ISP results

This function is called while camera is running for every frame processed by the ImgU, to process statistics generated from that frame by the ImgU. Algorithms shall use this data to run calculations and update their state accordingly.

Processing shall not take an undue amount of time, and any extended or computationally expensive calculations or operations must be handled asynchronously in a separate thread.

Algorithms can store state in their respective IPAFrameContext structures, and reference state from the IPAFrameContext of other algorithms.

Todo:
Historical data may be required as part of the processing. Either the previous frame, or the IPAFrameContext state of the frame that generated the statistics for this operation may be required for some advanced algorithms to prevent oscillations or support control loops correctly. Only a single IPAFrameContext is available currently, and so any data stored may represent the results of the previously completed operations.

Care shall be taken to ensure the ordering of access to the information such that the algorithms use up to date state as required.

Reimplemented from libcamera::ipa::ipu3::Algorithm.


The documentation for this class was generated from the following files: