Skip to content

GitLab

otb
otb
Open
Opened by Guillaume Pasero@gpaseroOwner2 of 5 tasks completed2/5 tasks

Define base classes for sensor models and implement the new models

Sub-issue of #1506

We should define base classes for sensor models (RPC, SAR, and external ones).

In order to implement progressively the new sensor models into the OTB, the work will be decomposed in 4 steps:

  1. Implementing an interface for sensor models
  2. Implementing the RPC model
  3. Implementing the SAR model
  4. Refactor the class otb::GenericRSTransform so that it uses ImageMetadata instead of ImageKeywordlist
  5. Implementing a mecanisme for the integration of External models (see #2039)

Here is a description of each step:

Interface for sensor models

Interface for all sensor models in OTB (RPC, SAR and external) and associated factory that instantiates objects from image metadata (RPC sensor model from MDGEOM::RPC ...). This can be done by refactoring the otb::SensorModelBase.

Example of class layout:

graph TD;
  otb::RPCTransform-->otb::SensorModelBase;
  otb::SARTransform-->otb::SensorModelBase;
  xxx::ExternalTransform-->otb::SensorModelBase;

To avoid the creation of Forward/Inverse duplicates, the transform direction can be a template parameter of otb::SensorModelBase, or even a class attribute.

The class otb::GenericRSTransform should be refactored to use ImageMetadata instead of ImageKeywordlist.

We could also add a small factory to instanciate the right transform:

  • MDGeom::RPC -> RPCParam -> otb::RPCTransform
  • MDGeom::SAR -> SARParam -> otb::SARTransform
  • MDGeom::SensorModel -> boost::any -> xxx::ExternalTransform (see #2039)

RPC model

The purpose of this task is to provide an API to GDAL's RPC algorithms (gdal/alg/gdalRPC.cpp).

This API will be supplied by the new class otb::RPCModel which inherits from otb::SensorModel. It will use the information contained in the image metadata dictionary (MDGEOM::RPC) to instantiate GDAL's algorithms.

RPC are usually used with a DEM. GDAL's RPC API doesn't allow to directly provide a GDALDataset as DEM. This prevents us to directly use OTB's new DEMHandeler. The adopted solution is to modify the DEMHandler so it loads the DEM as a VRT in vsi memory, because the link to the DEM in the vsi memory is accepted by GDAL's RPC API.

SAR model

The objective there is to port the current SAR model from OSSIM plugins to a new OTB class.

The new otb::SARModel class will inherit from otb::SensorModel.

Some test have been deactivated while implementing the geom file reading (!759 (merged)), because they rely heavily on the new SAM model. We need to reactivate them while implementing this model.

  • saTvSarDeburstImageFilterTest
  • saTvSarDeburstImageFilterTest2
  • saTvSarDeburstImageFilterTest3
  • saTvSarBurstExtractionImageFilterTest1
  • saTvSarBurstExtractionImageFilterTest2
  • otbSARAmplitudeEstimationImageFilterTest (DiapOTB)
  • otbSARAmplitudeEstimationTest (DiapOTB)
  • otbSARCartesianMeanEstimationTest (DiapOTB)
Edited by Julien Osman