Neural networks and people identification

Our company presents a system of persons identification in streaming video objects based on neural networks.

Case: school buses carry children around the area. The issue is in the lack of control over drivers and unauthorized passengers carried.

Intelli LLC introduced a solution based on neural networks. A video recorder with an LTE / 3G / Wi-Fi module is installed in the bus, which, in the presence of high-speed Internet connection, transmits the video stream to the servers of the GPSM data center. When the bus moves outside the range of high-speed Internet, data is recorded on the hard drive built into the DVR. As soon as communication resumes, the system provides access to the recorded data.

The first stage of the system implementation is online video surveillance with video broadcasting from the bus, the GPS tracking and driving style of the driver (control of aggressive driving).

A neural network analyzes the video stream and identifies people on the bus. After passing several cycles of training the neural network (usually 3-4 weeks), the system begins to determine the permitted and unauthorized passengers.

Below is a diagram of the system.

If the driver took not allowed passenger on the bus, the system detects the appearance of the stranger on the bus and generates a report on this event. Photos of the stranger, the place where the stranger appeared on the bus, the driver’s actions are sent in the form of a notification and report to the system dispatcher, who is obliged to initiate an investigation and report the incident to management.

The most important property of the system is learning ability. Deep learning leads to high accuracy in identifying objects of any complexity and overtime exceeds the ability to recognize objects of a human.