[X] My Assistant

[AdrianC]

Hello everybody.
I have been reading this forum for quite a while now (8 months+) but never got the nerve and data to actually contribute.
Let me start off by asking everybody two questions.

"What do you think about a high quality remasterization of MSL's MARDI descent and landing imagery?"

"Is it worth the effort having a higher resolution, higher frame rate version of the descent?"

Obviously i'm talking about a high quality remasterization (details will follow). My background is in digital visual effects (VFX) to be more specific 10 years in matchmoving, this involves advanced photogrammetry knowledge, and digital compositing.

Final results will be:
-a minimum horizontal resolution of 2.6K
-25 to 60 fps (TBD)
-derived telemetry overplayed on imagery and current stage (ex. heat shield separation, engines)
-original radio comms as sound.

I am aware of the fact that such attempts already exist (here). This is just a mere optic flow interpolation which dose not do any justice to the mission.

The simplified roadmap is:
-I will use HI-Rise data to create a digital elevation map then convert to a polygonal mesh
-Compute absolute position of each MARDI frame
-Camera project on the ground geometry a series of MARDI frames. Each temporal sample will contribute to a higher resolution image by stacking it on the ground geometry.
-Final rendering will be based ONLY on MARDI data. Interpolation between camera positions will be as physically accurate as possible.

This is no easy task, my estimate is about 40-100 hours of work minimum and lots of data missing.There will be some programming, some small tool developement. But since next month I will be out of a job (corporate financial problems) I thought I might ask you guys about the possible interest of this project.

PS If i posted in the wrong section, I ask the moderators to kindly help me.

[AdrianC]

Upon further inspection on what is available in PDS, I've hit a major snag. The raw data is a small subset of what was previously recorded in comparison with the whole JPG sequence found here > http://mars.nasa.gov/msl/multimedia/raw/?s...mp;camera=MARDI

This means from a total of 1283 frames recorded there are about 650 - 750 frames relevant for the EDL and on the PDS website, the latest release is only 388 frames long.
A quick look indicates that this release is showing every 3rd frame of what was captured. This means I am forced to work with the full JPG sequence in order to compute the whole camera motion.

About the raw non de-bayered images
I tried debayering blindly, without respecting the tables in the documentation. I have found that there are two schemes for interpolating the green channel but there is more work to be done here.
One conclusion is that the compression is applied, probably o board the craft for memory reasons. 4GB of DRAM if I remember correctly.
The point is that even the non de-bayered images are exhibiting signs of spatial compression thus reducing the color fidelity by introducing visible color macro-blocks.
Quick comparison in the attachment.

Attached thumbnail(s)

 

[mcaplinger]

...from a total of 1283 frames recorded there are about 650 - 750 frames relevant for the EDL and on the PDS website, the latest release is only 388 frames long.

Are you sure? A quick look at http://pds-imaging.jpl.nasa.gov/data/msl/M...EX/RDR_CMDX.TAB shows 1280 EDL frames (all as "MrdiRecoveredProducts") though these are inconveniently scattered across releases 1-3 because of their transmission times (32 on 1, 864 on 2, and 384 on 3).

One conclusion is that the compression is applied, probably o board the craft for memory reasons.

Per the documentation, the raw data are the digitized samples coming out of the camera and then passed through a 12-to-8-bit square-root table with no additional lossy compression. I'm not certain what you are concerned about in your images, but we've found that the two green pixels in the Bayer pattern have slightly different responses and this has to be accounted for for best color reconstruction. Bayer reconstruction is a very involved topic.