GunDetect - alerts you when it sees a gun
GunDetect - alerts you when it sees a gun
GunDetect sees guns with on-board image processing and alerts you in time to act.
GunDetect sees guns with on-board image processing and alerts you in time to act. Read more
About this project
GunDetect uses computer-vision processing to automatically detect a firearm and alert you via a text message. It may be used to protect your own gun from unauthorized use or to alert you to the presence of a gun during an invasion of your home or business. GunDetect is designed to compliment gun security. It doesn't require any changes to the gun itself. And GunDetect is completely controlled by you, the consumer. Your information goes to you alone, and you decide how to respond!
GunDetect may be set on a shelf or hung on the wall where it has a clear view of the area you want to protect. A standard electrical wall outlet and a Wi-Fi connection are all that are needed to install GunDetect. You may register your GunDetect online to receive automatic software updates and assign a phone number to which it will send alerts. No video is stored online for any version of GunDetect.
Other gun detection systems include:
- gun detectors that emit radiation to discover concealed weapons
- gun detectors for schools that work like smoke detectors, alerting officials only after a gun has been fired
- systems where "affordable" is between $15,000 and $60,000
Other firearm safety technologies have serious drawbacks for gun owners who worry about the reliability, delay and expense of technology that changes the gun itself - fingerprint recognition, bracelets, trigger locks, etc.
GunDetect avoids the pitfalls that have slowed the adoption of new technology for gun safety. GunDetect is affordable and is designed to complement gun use rather than potentially limit access in an emergency.
- Non-intrusive - does not emit radiation or store video
- Private – puts YOU in control of your information and how to respond to alerts
- Affordable - priced for consumers
- Preventive - early warning before accidental gunfire or intruder attacks
- Secure - an additional layer of security to guard against gun thefts and mishandling
- Complementary - GunDetect works with your other gun-safety measures
GunDetect comes in two versions, both of which are based on the latest computer-vision algorithms and optical sensing hardware. The difference is the location for the massive amount of number-crunching required to reliably detect a gun in an image.
GunDetect Premium is our main product and does all its vision processing locally using a powerful computing system that does not need to send any video data to the Internet – giving you the peace of mind knowing your private video never leaves the premises.
GunDetect Cloud has less local processing and uses our Internet servers to help crunch encrypted video data – potentially taking longer to detect a gun than GunDetect Premium.
Both versions of GunDetect wirelessly transmit the results – the presence or absence of a gun – over the Internet to generate a text message to a phone number you’ve selected.
Our software is already working and available on our web server (www.gundetect.com) for you to test before you back the project. Put in the URL for a stock photo of someone holding a gun or upload your own picture. If you register with your username and email address, you’ll have access to our fast servers for your tests. Help us improve our learning algorithms by clicking the appropriate button to indicate whether a person was holding a gun. We are training the vision system to ignore weapons that are not in someone’s hand – such as guns holstered, hanging on the wall, or sitting on a shelf.
We have also demonstrated our software on a real-time, functional prototype system, and you can see our first live demonstration in the videos. The prototype successfully recognizes guns and sends a text alert to a phone via Wi-Fi and an online SMS service. We continue to test and improve this system, but most of the remaining work involves the packaging for the production version. We’re still printing sample 3D enclosures and welcome feedback from our backers on the best industrial design to use for the high-volume injection-molded product. We expect to ship units to our early backers in about 6 months, and these early systems will receive the same software updates planned for the later versions.
(above video) Prototype v1 – Successful live demo of hardware and software
Our next step is packaging GunDetect into an attractive and functional form as well as refining its “deep learning” knowledge base.
GunDetect uses rapidly emerging technology called “Deep Learning”. The same technology is being developed for driver-less cars. The basic technology has been around for years, but only recently has the high-speed hardware needed to implement Deep Learning become affordable. With hardware now available, software is rapidly evolving, and our team is taking the lead in bringing this capability to the public.
Deep Learning software is “trained” by exposing it to a large number of images labeled as to whether or not they contain the object to be recognized. Once trained, GunDetect then analyzes new images to decide whether they contain the object of interest. Generally, results are improved when more images are used in training the software, so it’s fortunate that two-billion people now have cell phones that capture and upload images to the Internet.
We have the gun-detection software working with our prototype hardware, and we are evaluating some new vision processors that consume less power and would allow us to go into production with a smaller circuit board. We’ll be refining the physical package design over the coming months, since we plan to use many of the same industrial-design elements in both Cloud and Premium GunDetect. Software development is not gated by the hardware schedule, and we’ll release many of the software enhancements to the web-based demo for our backers to evaluate.
Risks and challenges
GunDetect is a product of nanoWatt Design, a technology solution provider for ultra-low power vision processors. The processor chips are fully proven in the field, and the printed circuit boards for first products are modifications of existing boards. The Deep Learning software used by GunDetect is fully under the control of nanoWatt Design engineers. Overall, the company expects no surprises in shipment and assembly of system components.
Getting to the point where we can ship GunDetect to your door involves the following:
1) Choosing the final processor board configurations. We are working with our suppliers during the design phase to minimize the number of components and to reduce the cost and size as much as possible while meeting our system requirements.
2) Arriving at a final form-factor. We are continually evaluating design variations to lock onto an enclosure that is both attractive and robust to protect the system over the long haul. Along with our electrical engineers, we are involving experts in industrial design, printed circuit board design, and plastics molding. We currently have excellent prototype designs for the two main versions of integrated systems that we are offering.
3) Setting up quality assurance. We have carefully chosen the highest quality vendors. We will also maintain an internal quality system to ensure that all products that are shipped meet our demanding standards. We are leveraging the depth of experience of our personnel, who have set up and managed many complex projects over their careers..
4) Finishing and testing all the pieces of hardware and software. We will anticipate the various use conditions and test to ensure that everything goes smoothly when you receive your systems. Our team members have decades of experience in building both software and hardware at the chip-design and printed-circuit board levels.
Some legal language: Before backing the GunDetect Kickstarter project, you should understand the technical limitations of the system, starting with the fact that computer vision systems offer no guarantee of safety. The GunDetect relies on visual images and cannot detect firearms concealed under clothing or any covering. It is not foolproof; it will not always be able to detect every weapon in its field of vision. Both false positives and negatives are possible, and indeed likely. As computer vision technology improves, advances and refinements may be incorporated into GunDetect’s software through over-the-air updates. As a backer of this project, you are helping to advance the state-of-the-art for the technology to make the world a safer place.Learn about accountability on Kickstarter
Support this project
- (36 days)