Though the number of robbery cases reported each year in the U.S. has come down to nearly half in past two decades and half, from more than 600,000 cases in 1990 to about 300,000 cases in 2015, most of these cases get reported after the perpetrators have fled the crime scene. Closed-circuit television (CCTV) or security cameras do help in bringing the culprits to justice but they fail to avert the crime from taking place, which often leads to loss of life and revenue.
Artificial intelligence (AI) may have a solution to not only sense a robbery but also alert the local police department before the problem escalates, according to the founders of Deep Science AI. Humanly tracking security cam is tiresome and inefficient, whereas deep learning models can spot objects such as masks and guns without any human interference and raise alarms. The company aspires to use the tireless analysis abilities of AI in order to bring in a cheap and yet reliable way to keep an eye on a store.
Summation of Footage from Thousands of Cameras
According to the CEO Sean Huver, Deep Science AI are among the first ones to construct a platform that can scale up to thousands of security cameras and have already began real-world deployment. The company has taken note of the fact that most of security cameras are IP-based, sending footage to data centers for archiving as well as monitoring in some cases, if paid. The system developed by Deep Science AI will sit on top of this footage stream, and is able to detect gun or face confidently with 30 pixels on target. In less assured cases, the footage are instantly sent to binary specialist network for reassurance, before human intervention is needed.
This system currently is capable of helping one analyst to deal with nearly 500 cameras at a time, and the company hopes to increase the number to 800 to even a thousand by refining their prototype in the near future.
Limited Deployments Already in Place
The concept developed by Deep Science AI to sense robbery and alert authorities is already functional at 18 locations in the U.S, running in a closed beta format for nearly a month. For instance, Sunoco, which has quite a few gas stations in the country, have expressed their interest in the cheap alternative developed by Deep Science AI.
Going forward, the ambitions of the company is to go beyond noticing guns and masks to detecting half in the air, brown window, fire starts, and smoke detection.