Funding for this project was canceled by the project creator on August 25, 2013.
About this project
The Pet Security System with the Virtual Microchip is the only identification system that is deployable after the dog is already missing.
This cutting edge system, already years in the making, employs facial recognition and breed identification for your dog in order to track every detail of your companion animal. Vet and medical records, ownership history, etc., are stored in a PSS (Pet Security System) which is accessible via an Android, iOS app, mobile web, or even in a regular desktop browser.
What is the Virtual Microchip?It is not a physical microchip by any means.
Instead, it is a collection of algorithms that can detect 350 breeds worldwide (all registries and a few mixes, we'll post an update to explain this in more detail), then identify your specific pet based on its unique facial scan. State and local partners in animal control,vets, or even the public could then check your pet against our system which would notify you if your companion has been found.
Ideally, you would have registered your pet before the little rascal got lost, but we can also add your companion to the system from your photographs even after your companion dog is lost.
The outside footage in the video was all filmed at Pet ResQ 911 (a non profit animal rescue in Madison, TN). Pet ResQ 911 works extensively with other good rescues, shelters, and agencies for dogs, including but not limited to the Nashville Humane Association and Metro Nashville Animal Control.
FYI, Pet ResQ 911 is the first adoption partner of Metro Nashville Animal Control since the new qualifications went into effect in June.
Metro Nashville Animal Control and the Tennessee State Health Department are set up for a demo during this kickstarter campaign! The entire state of Tennessee's animal control organizations are waiting with anticipation to see the system.
UPDATE: They are excited to see the demo, the director of Metro Animal Care and Control was sick. We will reschedule this demo when she gets well.
We want this system deployed across the entire United States and (with sufficient luck and support), Everywhere!
One more note on this topic for transparency sake. To be clear, this is not a Pet ResQ 911 non-profit project. This is a Pet Sec, LLC project.
Sounds Incredible. I want to see another dog!
I Also want to see some breed identification.. too!
No problem! Seeing is believing.
Although more dogs have been recovered after the 1970s on a year to year basis than before, that is only because there are now more dogs today then there were back then.
In terms of actual percentage of pets successfully returned, pet recovery hasn't improved since the 1970s or even the 1950s.
Frequency of Lost Dogs and Cats in the United States and the Methods Used to Locate Them For the direct download [pdf]
From the survey, "Unfortunately, both our estimated pet ownership percentage and estimate of lost dogs and cats within the past 5 years were lower than expected (39% vs. 50% and 15% vs. 30%). This resulted in fewer than 200 lost pets included in the survey. This seriously limited our ability to analyze less common responses including the role of identification tags in getting lost pets reunited with their owners." -- page no 311"
Some key points:
- The study never isolated pets that ONLY had a microchip and where no other method of identification was used a successful recovery.
- GPS collars aren't even in this study. They are simply not in the population enough to impact the study.
- Pet Tags and other visible items were the best object for animal recovery.
- Looking for your dog in your neighborhood as soon as your dog goes missing was the best recovery scenario.
What's wrong with the implantable Microchip. Doesn't that protect my companion just fine?
Rather than use up a lot of space on this kickstarter explaining the many pitfalls of implantable Microchips, read this knowledge base article from the AVMA (American Veterinary Medical Association)
Let's just say that there are plenty of incompatibilities, cancer, and movement of the chip to go around.
We can do better.
How does the Virtual Microchip work?
When you send us an image it goes to our API server, which then sends it off to our Job Queue to be processed. While in the Job Queue, this is where the magic happens. The image of the dog is ran through several layers of algorithms, first to validate the image isn't garbage (no dog), then to find what is interesting, then to compare against what it already knows.
**Any human faces or license plates in images are erased from the system. We are tracking dogs, not people!**
This system does not depend on a bayesian approach, but rather a "k-means, spatial histograms, and a novel configuration of SVMs with a novel arrangement/types of kernels, as well as several other algorithms" approach. I will be posting a more in-depth technical explanation to backers.
How do I load my dog into this system?
Since I couldn't remember Shania's name (the second Anatolian Shepherd in the breed detect video) we will show how to add her as a new dog to the system.
How Accurate is this system?
The over all accuracy is around 80%. I have several strategies to drive this number at least as high as 96% with a margin of error of 1%, but they all again require more servers.
Compare this to implantable microchips, which have a maximum 94.8% and low as a 66.4% sensitivity (READ: scan rate).
See this 2008 study: http://www.ncbi.nlm.nih.gov/pubmed/19046030
Why do we need funding?
In a word: Servers. Here's the rest of the situation
One more promise on data transparency:
We value open standards. We will always make your data available to you for exporting. The one and only exception might be the calculation we do on your image so that we don't give away the exact math that a particular kernel (or other novel algorithm) we are using to train our system with your pictures. However, we will export to you all human readable records at anytime without any hesitation.
Risks and challenges
Clearly this is a massive project. Like every software project of this scope, careful planning is needed. This system has been rewritten eight times from the ground up to ensure that when we get the servers/hardware we need that we can scale. All of the software has been written with parallelism and concurrency in mind. We are planning on doing a few significant refactorings after this project is funded. This should go according to plan, but could cause slight delays (think weeks added to the timeline not months).
With your support we will deliver the future of dog tracking and recognition.
Specific to International Backers:
In light of recent revelations in the media, the United States and American technology companies are under greater scrutiny than ever before. While we feel that this will ultimately be for the better, any backers living in countries outside of the the United States may live with a government that has the legal authority to block American technology companies from doing business in their country. Also, privacy laws are much stricter in the European Union than in the United States. We are unaware of any problems our system poses with EU law at this time.
We take privacy seriously and are transparent to the point that Andrew McElroy even posts his ham radio call sign on his twitter profile (aka you can find out in under 40 seconds where he goes to sleep at night). With this level of openness, it would not benefit Andrew to resell your data or meta data. Angry mobs are scary and best avoided by not making them angry in the first place.
Let me be very clear: We will not resell your data or meta data.
Thank you so much for your support.Learn about accountability on Kickstarter
Do you really need that much money for servers? How much computing power do you really need? I realize image processing is intensive work, but this seems steep.
Yes I really do. This isn't a toy project. I have to work through 7 "layers of algorithms"/stages -- machine learning can be very computationally expensive. I will post a more technical answer sometime this upcoming week to backers.
I already refactored to get out of matlab. Matlab alone wanted $30,000 just for the licensing of their software to use on a few server (30 jobs at the same time). Keep in mind that cost was just for software and not even for the servers needed to run it.
I wanted diesel out for two reasons.
1) I didn't have multiple dog in picture support turned on.
There are UI considerations and also it takes more computation to do this.
2) I haven't added diesel to the system yet to be identified.
The only money we will ever collect from this system are in the forms of two fees:
1) The initial fee to buy a PSS. We are thinking a one time $19.95.
We might cut a special deal for multiple dogs. say you have 5 dogs. the first would be 19.95 then we are running the numbers on if $10 for each additional dog is viable. It's looking promising.
2) A transfer fee of $9.95 to move a dog from one person to another. Only the person receiving the dog would pay that fee.
From our research, that's what microchips cost to transfer a chip.
We are open to lowering the cost of a transfer, but we want a payment of some sort because:
A) don't want dog theft/scams to work.
B) we want to somehow verify that the person receiving a dog is who they say they are.
Those are the only fees we have. That's it.
There is no cost to any animal control to use our system/send us pictures.
- (30 days)