Loss of visibility of an S1 image during a zoom out.
Mantis Issue 1352, reported by ludovic, assigned to gpasero, created: 2017-01-31
During zoom out on an S1 image, it will partially then fully disappear.
Open the S1 image available in OTB training course data package: https://www.orfeo-toolbox.org/packages/doc/Theia2016/
Then zoom back to 1:100 the image partially disappear. If you increase once again in 1:110 and more, the image will completely disappear.
This happen when an unchecked linear algebra occurs (singular matrix can not be inverted. Returning best estimation so far).
2017-05-02 12:18 - gpasero: Reproduced with a different S1 product, the image disappear when zoom level exceeds 1:120
2017-05-02 13:20 - julien: The thing is that sensor modelling for sar is not that stable if you request points that are very far from the image domain (which happens when you set zoom level 1:120). In that case the model should do some kind of linear extrapolation.
2017-05-02 17:26 - gpasero: I am trying to use the method ossim::SensorModel::extrapolate() but theBoundGndPolygon is empty in Sentinel1 model.
So I tried to fill the polygon manually at the end of the open() function, but it is not obvious.
Note : does someone know if this extrapolate() function is also valid for SAR sensors ?
2017-05-04 10:34 - gpasero: Calling extrapolate() function from worldToLineSample() doesn't work well (it generates cyclic calls).
There is an alternative solution which implies to know what is the sensor looking side : for every ground location that is closer than the near range distance or on the other side of the orbit, we compute the rangeTime differently to have a well defined and continuous transform.
This would be a significant change in SarSensorModel so I don't plan on integrating this modification in release 6.0