Picture from coverage by Adam Wilt at PROVIDEO COALITION:

Obvious questions that immediately come to mind:

• What is the life expectancy of an UAV? Unless it’s controlled by a trained RC guy, my guess is that it will crash within minutes of use.
• Insurance costs, anybody? Beyond destruction of valuable hardware, there is also a risk in injury for persons. Will even you get coverage?

More problems

• Vibrations can make photo and video capture unusable. Your UAV needs proper dampening and resonance cancellation features. I fear that cheaper UAVs don’t have such niceties. DYI solutions are not for amateurs, unless you have a structural engineering grade.
• In addition to controlling your UAV, you will need to remotely control your camera (zoom, aperture, …), so you probably need two people to operate it.

Aero Quartet Advice

Since our business is to repair corrupt video footage, we are seeing an increasing number of cases coming from UAVs users. Here is our advice about how to minimize risks and costs:

• Use cheap cameras. Image quality will likely be limited by vibrations and by your RC skills, rather than by camera. Crashes will be less painful.
• Use strong tape to prevent camera doors from opening. Battery, SD card can be ejected under a violent acceleration and jeopardize the take.
• We can always recover corrupt footage due to a “hard landing”, but the last few seconds cannot be repaired because they have not yet been written to the persistent memory card.

We have also covered “Destructive or High-Risk Video Capture” in a previous blog post.

Conclusion

I have no doubt that UAVs will become very common in the next few years, not just for video capture, but for many other activities. Today, this is still an emerging technology that is difficult to master, that presents some risks, and doesn’t have a legal framework. It’s not ready for prime-time, but as it opens amazing new possibilities, I’m convinced within a few years it will become a standard practice, just like use of camera cranes or steadycam today.

The good news is that we have managed to repair them with the same level of quality as damaged videos from the rest of Canon EOS family.

At bitstream level, there are a few differences: H264 video is now also using bidirectional frames (B frames), whereas Canon 5D Mk II and Canon 7D videos were only using keyframes and progressive frames (type I and P).

Audio encoding is the same, except for a 96 bytes padding found between audio and video. We haven´t found any convincing reason for the presence of the padding, but maybe Canon will use it in future models or firmware releases for advanced features.

• Mount a camera on a truck that will fall off a cliff
• Integrate a camera into a space rocket and shoot video until battery dies
• Mount a camera on a RC plane

Nowadays, it’s easy to find an inexpensive camera with good enough image quality and appropriate robustness. Even if the camera is crushed, the footage is safe because it is stored in the camera SD card, a small and robust component with no moving parts.

Just take the SD card out of the crushed camera and read the files!

Of course, the movies whose recording did not end “gracefully” are corrupt and have to go through a repair (and we know who can do this job…) but at the end you can have an impressive and valuable piece of footage that you couldn’t produce by “conventional ways”.

Disposable Video Cameras

Actually, the camera makers are starting to see a market for disposable video cameras, and it comes naturally as an extension of action cameras market, with popular models like GoPro HERO, Vholdr Contour or Drift HD. Established brands like Olympus have also shown interest, going as far as to develop a specific “Tough” series of shock-resistant and waterproof video cameras.

Don’t Try This at Home

If you want to record high-risk or destructive takes, you should plan it carefully:

• Fasten the camera to a strong support
• Try to avoid direct exposure of camera to shock or water. Pick position carefully, and use protective padding if possible
• Use strong tape to prevent camera doors from opening. Battery, SD card can be ejected under a violent acceleration and jeopardize the take
• And the most important: Never expect to capture the last second of life of a camera.

The Missing Last Second

Why? It hasn’t been written to the card yet when the camera dies.

Short Explanation:

In your camera, the recording process is more or less as follows: raw data from the sensor go to a memory “buffer”. Once the buffer is full (which corresponds maybe to one second of action), the encoding chip make his job and writes the compressed video in a second memory “buffer”. Once the second buffer is full (containing one to a few seconds of action), the data is written on the card. Only at this moment the data is persistent. (because the two memory “buffer” are volatile memory).

When the camera death occurs, the information in the two buffers (corresponding to last seconds of action) are in volatile memory. Not on the card.

Therefore, you should plan your take so that your camera stays alive for 3 to 5 seconds after the moment you want to capture.
If it’s not possible, a live recording system should be used, but that’s another story.

Canon EOS is “producing” an order of magnitude more repairs than other consumer cameras that outsell Canon 7D in the market.

Our job is to repair video files, so we are in a position to say something about it. Over the last two years, Canon EOS repairs have sky-rocketed, quickly becoming our main “product”. In most cases, a few clips shot during an event are corrupt, despite having recorded good clips before and after. The media data is present and is consistent (a condition to make the repair possible), but the movie index is missing.

It’s as if the camera had stopped working in the middle of the recording, or could not finish its job of adding the index just after recording ends. This smells like a firmware bug or system limitation. In any case, this is something that Canon can only hope to solve through a firmware upgrade — or — a new camera design.

(There are other failure modes found in Canon EOS cameras, like use of low performance CF cards, accidental deletion, battery issue … but those are the “usual suspects” that we find in all cameras. Here we focus on top Canon-specific failure mode: corrupt file due to firmware bug or system limitation).

FW1.2.5: Canon Acknowledges the Problem

In April 2011, Canon has released a new firmware, FW1.2.5, for Canon 7D.
According to Canon, Firmware 1.2.5 should eradicate corrupt movies:
“Fixed a rare phenomenon in which there were rare cases where movie files taken could not be opened when repeatedly shooting movies with specific CF cards.”

Strange wording. “Rare cases”, “specific CF cards”, look like words forced by PR or Marketing department, and in my opinion, try to masquerade the fact that Canon has only partially solved the problem (or the problems) and that you should not be surprised if failures keep occurring…

So I decided to dig into our repair history and figure out whether FW1.2.5 has changed anything over the last 6 months.

First, I notice that after FW1.2.5 release, our Canon repairs drop by 15%, then they start growing again, but at a slower pace than before.

How Firmware Roll-Overs Work

When a new firmware is released, all new cameras are manufactured with it. But it can take a few months until those cameras get into customers hands. If distribution and sales channel is slow, if inventory doesn’t receive the upgrade, the old firmware will continue growing in installed based for many months until new one first appears, then becomes significant.

Firmware upgrade is also possible on older cameras. But my guess is that only a small percentage of people is upgrading its camera firmware. One exception: if they are confronted to a problem, or receive advice to do so by Canon support, in a newsletter or in an Internet forum, they will upgrade.

FW1.2.3 Roll-Over as a Baseline

FW1.2.3 is transparent: It doesn’t claim to fix any failure mode causing corrupt clips.
That’s why it’s a good baseline to see how firmware roll-overs happen in Canon EOS world.

We look at the picture 6 months after the release to give it time to significantly affect the installed base through new sales and manual upgrades. The chart shows how repair cases are distributed by firmware versions:

• After 6 months, the latest firmware accounts for 18% of repairs.
• FW1.2.2, the last one, is now the most represented. This is due to inertia of installed based and sales channel.

Now with FW1.2.5, a Different Story

The big change with FW1.2.5 is that it claims to solve clip corruption issues. So all other things being equal, we will evaluate the efficiency of the firmware fix by comparing the occurrence of repair cases.

If FW1.2.5 were not fixing the bug, then 6 months after release, the distribution of repair cases by firmware releases would look exactly like the FW1.2.3 chart (see above).

It turns out that it’s NOT the same:

The significant differences and their possible explanation are:

• Volume of repairs decreases by 15% (as noted in first chart). Sounds like there is improvement1
• FW1.2.5 represents only 12% of total, which is significantly lower than 18% of FW1.2.3 six months earlier.
• Considering growth of installed base (with important FW1.2.3 inertia and FW1.2.5 in most recent sales), FW1.2.3 should be the most frequent version, but its volume is almost the same as six month earlier!
• We see an important and unexpecte decrease in FW1.2.2.

I can only understand those results by imagining that customers have upgraded en masse from FW1.2.2 and FW1.2.3 to new FW1.2.5. This is probably due to Canon pushing FW1.2.5 as a “problems killer”, and due to Canon users advocating for upgrading firmware in Internet discussions.

But then, FW1.2.5 should now the most frequent in the Repair Cases. And actually, it’s the less frequent one!

This can only be because FW1.2.5 is producing significantly less corrupt movies than the previous versions.

Conclusion

To all Canon 7D users: Please upgrade to FW 1.2.5 and do it fast or you will have to use our Movie Repair Service!

This week, we will see how we can improve and try to salvage all the footage anyway.

SD and HD

Let’s go back to our disk “disaster”: We have used a data recovery utility and we have files but they do not contain what we expect:

On top of that, the files were recorded in two formats:

• Biking and Party in high definition (HD) and mono-aural audio,
• Birthday and Concert in standard definition (SD) with stereo audio.

And a different repair kit is needed for SD and for HD, but the real problem is that our recovered files can contain both formats mixed, like “Party”.

What will happen?

You will have to try each file with each kit:

• SD kit on an HD file will produce an empty file, or garbage video.
• HD kit on an SD file will do the same.
• On files with mixed footage, the kits will only repair the corresponding segments of the file. “Party” will give two repaired files, an HD file with some biking and an SD file with part of the birthday.
• “Birthday” file will give empty files with both kits
• None of the repaired files with contain an event end-to-end, due to recovered files not matching the media data. No party footage will be recovered at all (“Party” file contains biking and birthday)

Frustration

Bottom line: You will spend a lot of time repairing every file with every kit and you will end up with a mess:

• 80% of biking but in two separate files (one of them named Party-0.mov)
• Only 50% of the birthday recovered (in a file named Party-1.mov)
• No party footage found
• Only 95% of concert recovered. A couple of songs missing

Imagine if you have to repair a 2TB hard disk with 1000 files and 5 kits!

DeepMediaScan can do better than data recovery

Enter DeepMediaScan, a free feature inside Treasured.

The problem we have is to map correctly the data regions inside the disk, a task at which the data recovery utility failed.
DeepMediaScan helps you do that by creating a Disk Map.

Just like archeologists to put a grid on the ground area they are searching, DeepMediaScan puts a grid on your disk, and divides it into tens or hundreds or cells:

DeepMediaScan lets you specify the number of cells. Sometimes a fine-grained map is needed, in particular if disk is big and files are relatively small (needles in a hay stack).

But the idea is simple:
For every cell, DeepMediaScan will generate a preview picture, that tells you the format, resolution, and visually indicates what clip it belong (biking, partying, birthday and concert images are easily distinguishable).

The benefits are clear:

• Now you know where you footage is stored, with the resolution of a cell
• Thanks to format and resolution information, you know what kit is needed for a given cell or group of cells
• On big disks with hundreds of clips, you can focus just on the project that you need to recover, you don’t have to blindly try all files on all kits as above

All in all, with the visibility brought by DeepMediaScan, a big recovery becomes manageable.

What about A6?

Some cells, like A6 or B5, contain media from 2 different clips and formats.
In the Disk Map, however, you only get one preview image, corresponding to the left-most clip/format: A6 will show biking and HD format.

To minimize loss, you can either increase the scan resolution (smaller cells), or consider the A6 cell as disputed between biking and birthday (since A7 coming just after is a birthday image), and schedule A6 to be repaired in two different clips (biking and birthday).

In the third and last part of this story, I will explain how to do it and more…

• doesn’t contain stuff corresponding to the file name or extension,
• contains footage from several clips,
• or contains large swaths without footage, or are completely empty

And in this case, your confusion is legitimate, because after a storage failure, we have in a twilight zone as far as data storage is concerned: Normal assumptions are no longer valid.

In this post, I will try to explain what is going on “behind the scenes” and how the repair results can be optimized.

Disk, Data and Files

Through this serie of posts, I will use the terms “disk”, “data” and “files” very often, so let’s start defining what I mean and how I will represent them in the diagrams:

disk stands for physical storage, it can be your computer hard disk, an external hard disk, an SD memory card, or and USB pen drive, for example. It is represented by an area whose surface is the capacity.

data is represented by a colored area. For example, green is used for the biking video. Note that the dark green region at the right side represents the index of the video, whereas the light green is video media. Both are necessary for the video to be playable.

file is represented by a boundary and a bubble with the file name.

In a normal situation, your disk contains several files and looks like the diagram below:

• We have recorded 3 video clips: Biking, Joke and Party
• File boundaries match exactly the corresponding data
• Disk is getting filled from left to right, in rows from top to bottom.

Joke, not so funny

Now imagine that we don’t find the joke very funny after all, and we delete the video file.

Note that only the file has disappeared, the data is still written on the disk, but unreferenced. This is why recovering deleted files is possible in the first place:

Later we record more clips called Birthday and Concert. Let’s observe the new diagram:

• Birthday file has overwritten Joke data
• Birthday is one file but occupies two regions: We call that fragmentation.
• Concert is stored in one region, because line-breaks do not count as boundary.

Bottom line: Unreferenced data will be overwritten when new files are added. So whenever you experience a storage failure or data loss, stop using the disk immediately. Reading doesn’t harm, but writing will end up wiping definitively your lost data. Now we can’t recover Joke because Birthday is written on top of it.

IMPORTANT: When repairing data or video from a disk, always operate from a different disk!
This avoids writing to the damaged disk.

Unplayable videos

To be playable, a video file must contain media data (video and audio), an index, and be coherent.

• Without media data, the file is not just unplayable, it’s not repairable either.
• A file with media but without index can be repaired.
• A file with media and incoherent index can be repaired, but is unplayable.

The most typical corrupt videos are those that you find after a recording failure. The media data is present, but unlike a well-formed video file, the index is not here:

This is because the index is written at the end of recording, and if we experience a failure (battery dead, camera overheating, …) this will never happen.
Note also that those unplayable files can also contain some empty space at the end, where data was about to be written when recording stopped.

Disaster strikes!

Now the dreaded storage failure suddenly happens. We no longer see any file in the disk (or we have files with funny names, or files that are now unplayable).

The first thing to do is to quarantine the disk, to avoid writing on it. Because the data is still here, waiting to be recovered, but it is also vulnerable since it is unreferenced.

At this point, many people use a free or commercial data recovery utility to “undelete” the files.
The result is usually disappointing, because video files are difficult to recover due to their diversity, complexity and huge size.
You often end up with a few good files, and plenty of bad ones: (see diagram below)

• File boundaries do not match exactly the data regions
• “Biking” contains only half of the footage and is unplayable because index is not present.
• “Party” now contains footage from Biking and Birthday, and is unplayable for being incoherent.
• “Concert” is unplayable, because the missing bit makes the whole file incoherent.
• “Birthday” is empty!

Note that I’m using quotes to differentiate the recovered files (“Party”) from the original ones (Party). Indeed data recovery often mixes file names and contents, so after recovery, file name should always be taken with a grain of salt.

File sizes, however, are always a good indicator of what you can expect. A 160 kb file is too small to contain footage. A 16 MB file can contain a few seconds (or can be empty). A 160 MB file can contain several minutes of footage, …

We can also find combinations of the examples above, like a recovered file with footage from 3 or more clips, a recovered file with large empty regions and footage, …

Repair will give limited results

Even if you invest time (and money) in repairing those files, you will probably not get a satisfactory result. As we have seen before, the recovered files do not contain all the missing footage, and can also contain footage from several clips:

• Once repaired, “Biking” will just contain half of the original video.
• Once repaired, “Party” will contain some biking and part of the birthday video!
• Once repaired, “Concert” will now have the first songs.
• “Birthday” will not repair, it contains nothing!
• Party cannot be repaired. Parts of other clips are still missing.

But there is hope…

To achieve optimal results, we would need to recover files that match better the data regions, and then repair them.
Fortunately, this is possible, using our advanced repair tools, and I will explain it in the second part of this post.

But now I will have to add a third category: counterfeit cards.
There is strong evidence that counterfeit cards are spreading like plague since second half of 2009. Usually sold through Internet, shipped from Asia and significantly cheaper than equivalent cards in retail stores.

You won’t notice until it’s too late

Unfortunately, you won’t notice it until it’s too late: the card works fine, then fails silently. One day, you will want to copy or watch the movies and you discover that the majority of them are unplayable.

Treasured, the diagnostic app for damaged movie files, will give some hope: Unplayable files indeed contains some video, and you can see the preview. But you will also notice that all the unplayable files seem to contain the same few seconds of video. This is the symptom of a counterfeit card.

Here is why:
The 16 or 32GB card that you have purchased on this obscure website is in reality a 2GB card. It masquerades its capacity to look and feel like a 32GB file, and at the beginning, while you are filling the first 2GB with data, everything is fine.
But once you go over the real capacity, any data that you write to the card is lost. The card doesn’t complain, the operating system doesn’t notice. Everything seems to work, but in reality your movies and photos are never written to the card. Because it’s already full!

Needless to say, such unplayable movies can’t be restored. It’s possible to repair the content of the card, but not what was not written in the first place.

Please take a look at those web pages to learn more:
http://sosfakeflash.wordpress.com/
http://www.overclockers.com.au/wiki/Fake_Memory_Cards

and to this utility to verify your card:
http://sosfakeflash.wordpress.com/2008/09/02/h2testw-14-gold-standard-in-detecting-usb-counterfeit-drives/

Similarly, every two or three months, I receive a movie repair that is at the edge of impossible. It has this rare quality of being too damaged to be repaired with known techniques, but at the same time I foresee that there is a small chance that a new technique may fix it.

Officialy, I don’t call it Everest-class, but rather Investigative Repair, to convey to the customer three important ideas:
- I have to develop something radically new. A unique solution for a unique case.
- it’s gonna be expensive (I charge usually one order of magnitude more for those repairs)
- at the end, I can come to the conclusion that the repair is impossible. Results are not guaranteed.

I do such repairs primarly to challenge my repair ability and to push the enveloppe of the discipline. As boring as repairing movies may seem, an Everest-class repair is when there’s adventure, struggle and achievement.

The last Everest-class repair came last month: Canon EOS 5D Mark II files with some bitstream corruption. Mr Szot, the polish director, did not notice the problem until the production was finished. The clips, needed for a 15 minutes short movie called “An Exclusive”, could not be shot again.

I observed around fifty defects in a dozen of files. The footage is encoded in H264, where one wrong bit can propagate nefarious effects over a dozen of frames. Even with “creative editing”, it would be impossible to tell the story: For some important takes, the footage was unusable.

In 99% of the repairs, the problem consists in re-indexing the clip: Audio and Video media is fine but the “table of contents” that tells where the data for each frame, is missing.
But here, it’s the opposite. The table of contents is fine, but the media is corrupt.

Not that corrupt, according to real world standards, since the amount of bad data is only 1 for 100 millions. Like a rare disease that would affect just 70 people in the world population. The typical needle-in-a-hay-stack problem. Change needle for bits and haystack for a file with one billion bits and you get the idea.

Bits Flipping Party

The repair technique lies on a simple idea: when decoding the frame, the wrong bit will cause errors and exceptions that eventually stop the decoding process. That is what we observe in a defective frame: the top is ok, then something occurs, then the rest of the frame is not decoded.
The location where it stops should be close to the wrong bit. Never before, at a short distance after.
From here, we will go backward and flip the bits one by one, until we get something that decodes without error.

I made a bold assumption here: that wrong bits are extremely rare. So rare that we can consider that there is only one wrong bit involved in every decoding error.
I had nothing to really back this assumption: If it’s true, we would be able to repair. If it’s wrong, it would be impossible. Checking the assumption thus became a priority before engaging into time consuming developments.

There’s a second assumption: The distance between the wrong bit and the error is small.
One iteration will be needed for every bit, and iterations are slow since they decode several video frames each. Not just one frame, but the whole group from the I keyframe to the damaged frame. That can take several seconds, so if we have thousands of iterations, it would become impracticable.

Prototype and Automation

I assembled a prototype from various pieces, an open-source H264 decoder, a couple of small programs to flip bits and to detect errors and generate pictures from potentially successful iterations.
I tried it on a first defect, a slow process since almost everything had to be done by hand, and results have to be carefully interpreted.
When I finally managed to get a good picture out of the prototype, it was unbelievable. I had found the needle.

Since I had almost fifty defects to fix, I spent some time automating the process. At the end, I would only have to send a few command-line commands, choose manually the starting point of the search, and launch it.
The computer would run for minutes, sometimes hours, until it starts spitting pictures.
I had to review the pictures one by one, until I could find one that was perfect.
Then, I would modify by the wrong bit in the movie and verify that it fixed the frame and also the rest of frames in the “Group of Pictures”.

In some cases, fixing a frame would just unveil a defective one a few frames later. Just as one train may hide another…

Finally, I managed to clean completely all but one defect. This was quite a surprise. I would never have anticipated such a desperate repair attempt to work with a 98% success rate.

My customer, Mr Szot, is happy. His film “An Exclusive” will be presented this week to the polish public.

For Windows users, I have recently found a good technique to create a disk image from the damaged disk or card, which is the first step towards recovery of footage.
Note that this can also be useful to Mac users if the disk is not visible in the Finder, but is visible on a Windows PC.

In a nutshell, we use a small utility called dd for Windows that is able to read the raw data of a disk. (From now on, I will just use the word disk, and it can refer to a hard disk, a memory card or a USB drive)

Under the premise that the disk is readable, ie Windows detects it when you plug the disk or insert the card in the reader, “dd for Windows” should be able to create a file with about the size of the disk.

Here is the process, step by step:

http://www.chrysocome.net/downloads/dd-0.5.zip

This program is free and distributed under GPL license. For more information, refer to the product page.

dd --list

And you will get a list of the devices connected to your computer.
Here is what I get for example:

rawwrite dd for windows version 0.5.
Written by John Newbigin
This program is covered by the GPL. See copying.txt for details
Win32 Available Volume Information
\\.\Volume{aef46cf9-3e3d-11de-b8c6-806d6172696f}\
link to \\?\Device\HarddiskVolume3
fixed media
Mounted on \\.\c:

\\.\Volume{aef46cf8-3e3d-11de-b8c6-806d6172696f}\
link to \\?\Device\CdRom0
CD-ROM
Mounted on \\.\d:

\\.\Volume{6f41f4b2-d11a-11de-b318-001d4f88486c}\
link to \\?\Device\Harddisk1\DP(1)0-0+5
removeable media
Mounted on \\.\f:

Don’t be afraid, it’s just a list of the devices connected to your computer.
Here the interesting device is the F:\ volume described as removeable media: this is the card that I want to recover. Take note of the corresponding volume name:
\\.\Volume{6f41f4b2-d11a-11de-b318-001d4f88486c}

This volume name starting with \\.\Volume{ and ending with } will be used in step #6.

dd if=\\.\Volume{6f41f4b2-d11a-11de-b318-001d4f88486c} of=c:\usb.img bs=1M

This command creates a file called c:\usb.img that is a carbon-copy of the damaged disk.
The operation can take a long time. For 1GB, it can take one minute or more. For 120GB, it will take several hours.

Note that you can monitor the size of the c:\usb.img file as it grows from 0 to the size of the disk: It will give you an idea of the progress.

What do I do with the .img file?
Now if everything went fine, you should have a .img with a size similar to the disk capacity.
If the disk still contains footage, the footage must now be inside this file also, and we can do a diagnostic of the file and hopefully, repair it.

For Mac users, Diagnostic and Repair have already been covered in a previous post.
For Windows users, the .img file can now be used to create an extract that will be uploaded to Aero Quartet. This is explained in the Movie Repair Guide in the Windows page.