AlgoLook, a company specializing in artificial intelligence products, is patenting a method that enables video cameras to detect and identify weapons.
The technology may involve a weapon detection application, which includes a set of programs. It processes each frame captured by one or more video cameras. It includes a weapon detector worker, which itself comprises three components: a detection model, an image extractor, and a match classifier.
The weapon detection model extracts data from a video frame, which is fed into a convolutional neural network. The network outputs the detection of a weapon within the image, and may also identify where the object is in the frame.
The weapon image extractor may crop the edges of the video frame down to the location of the weapon. Its output is sent to the weapon match classifier, which performs a second look to determine whether a weapon is present in the video frame. It then accesses data organized via a vantage point tree algorithm. Example armaments are classified under trees for pistols, rifles, and holsters.
Instances of objects being located and classified are logged under an event queue. A separate program generates alerts for these events. These alerts include the weapon image, the object’s classification, the camera that captured it, and the time and date of the event.
An example of the system processing images in accordance with the disclosure.
AlgoLook says the technology may also include a face detection application, called the face detection worker, comprising programs that similarly detect, image extract, and classify facial data. Its classifier may organize information under different types of trees, depending on where the system is installed.
The patent application states that at a business, for example, data may be sorted under an executive tree, an employee tree, a contractor tree, a visitor tree, and an unknown visitor tree. This approach could help security personnel locate potential threats while eliminating guesswork.
AlgoLook has a stated goal of developing AI for security and safety applications, and its patent application stands to reconcile two major, controversial aspects of the modern world: the global populace owns over a billion guns and is under surveillance by over a billion cameras. The technology proposed in the patent application may prove useful in simultaneously keeping guns in locations where they are illegal and boosting the utility of mass surveillance systems.
The featured patent application, “Weapon Detection and Tracking”, was filed with the USPTO on Dec. 31, 2020 and published thereafter on July 8, 2021. The listed applicant is AlgoLook Inc. The listed inventors are John Dankovchik, David E. Harrison, and Jeremy C. Patton