Distributed Computing to help in search operations

[socaldmax]

It always saddens me to hear that a vessel with her crew has abandoned ship somewhere on the ocean, I always hope for a safe rescue, but I'm still surprised that in this age of technology, some people sre never found.

In the past, I've been involved with distrbuted computing efforts such as SETI@Home (search for extra terrestrial intelligence) with my computing scores from my 6 desktops counting towards Team ArsTechnica Lambchop. Later, my 2 i7 desktop computers were dedicated to Folding@Home sponsored by Berkeley, which used specialized software to not only harness 100% CPU power, but also used your GPU to get every last Work Unit crunched in an effort to crack the code to Alzheimer's and other neural disorders.

Recently, a R&C Leopard 44 crew lost contact back on Jan 18 and a search was conducted. In addition, a crowd sourcing effort was undertaken on Tomnod to use volunteers to visually search for boats or liferafts on sections of open ocean.

It occurred to me that if the same software platform used in distributed computing was used for searching for boats, liferafts or debris, it would be thousands of times, if not millions of times faster than the current visually scanning sectors. The server could send out sectors of ocesn for your computer to scan, you use a small software app specifically written to search for objects on the ocean and it can magnify and enhance the images and discern between whitecaps, clouds and actual boats or liferafts.

I did some online searching and couldn't find any distributed computing search projects currently operating. I've been very impressed with the OpenCPN project, an amazing amount of work has produced a fantastic product. I was thinking there are many talented programmers in the cruising comunity, and was wondering if there was any interest in starting a distributed search project in cooperation with Tomnod, using their images and server to send and receive sections of ocean for each computer to work on.

One could search for a specific vessel, or lend their computing power toward the search for any vessel which has lost contact or overdue, the server would track which sections of ocean were sent to each volunteer computer and send promising object sightings to MRCC authorities to help guide them in their search.

I know it's all technically feasible, and probably a lot easier to do than other projects, since there are many distributed computing platforms already in use, the question is, is anyone interested in working on such a project for what I feel is a very good cause?

[hellosailor]

I expect the problem is that the distributed software would have to be capable of optical recognition on each "tile", evaluating debris and whitecaps, etc. and trying to discriminate them versus what a "boat" might be.


That kind of optical recognition was highly classified military programming back in the 1980's, I'm not sure anything so sophisticated is available on the public market today. But if you started with folks who are currently doing AI research and pattern recognition, megabuck companies who are refining this technology so their robots can tell the difference between ripe strawberries and bugs (honest)...you might get them to work on the critical image recognition portion, and then the rest could be a piece of cake.
Remind the big corporations, this is a tax deduction, great PR, and a great testing tool for their programs. It makes $$ and sense for them to join in.
The various intelligence agencies who are using similar programs today, won't even admit they are doing so, much less share them with you. Which is only to be expected.

[socaldmax]

That's a good idea for possible sources of software.

I was also wondering if NOAA or some other related agency might have already written the software, possibly in an attempt to discern wave size, or waves from clouds, etc. It shouldn't take too much tweaking to set object size, shape or sharpness thresholds to look for boats or liferafts.

[hellosailor]

If it was that easy, you don't think someone would have made use of it already? In the search for the Air France flight that went down? Or the Malaysian Air flight they still can't find a trace of?


I suspect the classified aspects will keep any vaguely governmental options removed from the public market. But corporate AI and robotics...Honda, IBM, Google, Mercedes (and all the self-driving car interests)...there are lots of corporate players in this area, without those constraints.

[ShaktiGurl]

It is an interesting idea. There are many programs that perform anomaly detection and feature extraction on imagery - some classified and some commercial. These algorithms can and have been used for the detection of vessels on water surfaces. The methods mostly involve shape/texture analysis.
I believe much of the imagery delivered to Tomnod (from DigialGlobe) has been through these algorithms before being passed to end users to visually inspect them because automated methods still only detect so well and there is often a manual process required for review. DigitalGlobe is already playing in the distributed computing (and cloud) world if you are thinking along similar lines.

[socaldmax]

Originally Posted by hellosailor

If it was that easy, you don't think someone would have made use of it already? In the search for the Air France flight that went down? Or the Malaysian Air flight they still can't find a trace of?


I suspect the classified aspects will keep any vaguely governmental options removed from the public market. But corporate AI and robotics...Honda, IBM, Google, Mercedes (and all the self-driving car interests)...there are lots of corporate players in this area, without those constraints.

Yes, I think it's pretty easy. Considering there are some very low priced video surveillance systems that include motion detection, I don't think it would be too hard to write some software that could pick out an object the size of a boat or life raft from a predominantly dark background.

Considering that the US Coast Guard is in the SAR business and gov't agencies are cooperating with each other much more lately (Navy and CG doing drug interdiction with DEA, BLM and local sheriffs enforcing state and federal laws, DoN assets building CBP border video surveillance equipment via DHS contracts, shared satellite imagery between agencies) I don't think it's too far fetched at all for the CG to get some assistance with image recognition software that isn't classified.

Perhaps they're already on their way to implementing it, but not publicizing it, much like their future use of drones to help in open ocean searches.

[socaldmax]

Originally Posted by ShaktiGurl

It is an interesting idea. There are many programs that perform anomaly detection and feature extraction on imagery - some classified and some commercial. These algorithms can and have been used for the detection of vessels on water surfaces. The methods mostly involve shape/texture analysis.
I believe much of the imagery delivered to Tomnod (from DigialGlobe) has been through these algorithms before being passed to end users to visually inspect them because automated methods still only detect so well and there is often a manual process required for review. DigitalGlobe is already playing in the distributed computing (and cloud) world if you are thinking along similar lines.

That would be great news! Do you have a link to any info on who is doing this?

[ShaktiGurl]

ENVI is a software package that performs feature extraction, anomaly detection and general image processing (exelisvis,com). There is a scalable enterprise solution called ESE that supports distributed computing and now we are moving to cloud based solutions with DG and others such as Airbus:

http://www.geo-airbusds.com/en/4445-...ction-tracking

Note how they are using primarily AIS as the primary means of finding ships. The detection and extraction of objects on water is not as simple as it sounds with the type of imagery that is available worldwide, at the right time, cloud-free, and at scales appropriate for detection of objects of the size of most recreational boats.

[socaldmax]

Lots of great information there, it will take a little while to look at all of it.

Thank you!

[hellosailor]

socal-
" I don't think it would be too hard to write some software that could pick out an object the size of a boat or life raft from a predominantly dark background."
The problem is way more complex than "white on black". Heck, WIRED ran a photo in 2013 or 2014...someone noticed a rather large construction area in China, and for months they were trying to figure out what it was. AFAIK no one ever did, and that's an image of a large ground site or facility. You couldn't tell if there were ditches or buildings--both look alike depending on the lighting angle. Irregular objects at sea, versus whitecaps and reflections? Granted the available images are getting better and better (higher resolution) and sometimes multi-spectral are available (rarely to civilians), but the folks who really have experience with this? Are still military photo analysis intelligence folks. And their first rule is, understandably, don't let anyone know what their capabilities are. So...you'd probably have to try someplace like the JPL, where they have had great image analysis programmers, who might be allowed to do pro bono work. Anyone can program--but really sharp programmers? Not so easy to find! And the USCG's budget for anything, is simply one more thing they don't have the funds to search for. [sic]

[wellin]

With the extreme size of water and very short line of sight i cant see this helping more than a millionth of one percent in coverage. If that.

Sent from my SCH-I415 using Cruisers Sailing Forum mobile app

[Pie seas]

If it was easy, someone would have done it already. This is a great idea and with the rate of computing development increasing every moment, it can and will be done. Googol earth can show me how many chooks were out on the day they scanned my place. We have the technology, lets develop the skill.

[hellosailor]

"on the day they scanned my place."
That's the other catch. Google Earth may be 2-3 years out of date, more in some areas. Getting high resolution photos from satellites as they happen to pass over an area, costs. Getting the best photos by positioning a satellite where you need to task it, costs way more.
What's available in imagery that is taken daily? Or every other day? At what cost?

[Pirate1221]

Originally Posted by Pie seas

If it was easy, someone would have done it already.

SOCAL---

Don't let the naysayers and "Debbie-Downers" on here discourage you. I think it sounds like an awesome project. Enough time, desire, and effort, I'm sure you can make it happen!! Wish I had the particular talents needed to help!!

Always here for encouragement, though!!

[socaldmax]

Originally Posted by hellosailor

"on the day they scanned my place."
That's the other catch. Google Earth may be 2-3 years out of date, more in some areas. Getting high resolution photos from satellites as they happen to pass over an area, costs. Getting the best photos by positioning a satellite where you need to task it, costs way more.
What's available in imagery that is taken daily? Or every other day? At what cost?

There are a large nuber of satellites, including the Spot series, Pleiades 1A and 1B, and a variety of others. Pleiades alone has very impressive specs:

Agility for Responsive Tasking
This agility coupled with particularly dynamic image acquisition programming will make the Pléiades system very responsive to specific user requirements. Individual user requests will be answered in record time, thanks to multiple programming plans per day and a state-of-the-art image processing chain. Performance at a glance:
Image acquisition anywhere within an 800-km-wide ground strip with 70 cm of resolution
Along-track stereo and tri-stereo image acquisition
Single-pass collection of mosaics (strip-mapping) with a footprint up to a square degree
Maximum theoretical acquisition capacity of 1,000,000 km2 per day and per satellite

I saw an image where I could clearly see docklines in the pic. That's all weather dependent, but still, the capabilities are impressive.