Kola NPP (Vizorlabs Platform solution of video analytics)

2019: Creation of a system of the video analysis of observance of requirements of the security regulation and industrial security

Automated system of the analysis of a video flow from surveillance cameras in production premises with the high-voltage equipment where observance of requirements IS (industrial security) and TB (security regulation) is required. Control of 25 types of violations of TB and PB is implemented, expansion of a scope of a system is planned.

To system implementation on the NPP about 80 violations of PB and TB a week were registered. After system implementation the number of violations decreased by 10 times – to 8 in a week.

The system of intellectual video surveillance completely automates process of detection of violations of TB and PB and replaces with itself control of the image from cameras by the worker manager who quickly is tired and loses concentration of attention when viewing a uniform video series. A system registers 95-98% of violations of TB and PB and allows to prevent accidents in real time.

Surveillance cameras in production premises fix all movements of personnel and a work progress. The neural network analyzes the image from cameras, detects violations of TB and PB, and transmits a signal of violation to the shift supervisor's monitor. The shift supervisor on a radio communication gives a command about suspension of works before elimination of violation of TB. A system can automatically send commands for a stop and continuation of works.

The business effect of system implementation consists in reduction of direct expenses on control in the form of the salary of managers and reduction of losses from idle times of production because of injury rate.

The minimum feature set of observance of TB and PB detected by a system:

• 1. Wearing helmet and podborodny belt
• 2. Carrying and provision of a protective guard it (is lowered, lifted, partially lifted)
• 3. Wearing special gloves, trousers, footwear
• 4. Overalls are completely buttoned, sleeves are not rolled up
• 5. Distinguishing of several types of overalls
• 6. Tracking of provision of the worker concerning the boards KRU

A system represents a cluster of neural networks or detectors which carry out certain tasks. Detectors in a system can be combined and cascaded, transferred results of one detector to another.

The first detector looks for the objects similar to the person in a frame, the second defines his belonging to this or that group of personnel as overalls, the third looks for objects of SIZ and violation in use of SIZ obligatory for this employee group.

For detecting SIZOV the complex detector developed by VizorLabs based on detector open source is used. The detector is capable to work at a video flow of low resolution (640*480) and to detect the SIZ type, existence of a helmet, a guard status, existence of uniform trousers.

For provision of the employee of rather dangerous zones and provision of extremities the detector which is independently developed by VizorLabs – the skeleton composer is used.

The module of training allows any specialist who is able to use the computer, in addition to train neural networks for detecting of the SIZ new elements or to teach to distinguish a system new type the specialist of clothes, using information which arrives from cameras.