Perm-based IT company TrafficData trained a neural network to recognize accidents based on CCTV footage

Posted by

The Perm-based company TrafficData has developed a traffic accident recognition module based on CCTV footage. The development detects the signs of an accident on video with 90% accuracy and signals the dispatcher about the incident. This was reported to CNews by representatives of TrafficData.

The development consists of two components: a mathematical algorithm and neural networks. The algorithm evaluates the stopping time, the speed and direction of vehicles, as well as the situation on the road and the reaction of people. The solution of the Perm developers collects statistics on how long the cars are in order to determine the duration of the abnormal stop. It is calculated separately for different sections of the road, which is important, for example, for intersections, where different stopping times are allowed in different directions. The TrafficData algorithm dynamically updates data on the current traffic situation, and in the event of a traffic jam, it recalculates the allowable stopping time for cars.

After the algorithm has identified an abnormal stop, neural networks are “turned on” to check whether there are signs of an accident: a collision and damage to a car, downed pedestrians, fires. Neural networks help to reduce the number of false positives in the system. After an accident is detected, the signal is transmitted to the dispatcher, which allows you to quickly respond to the incident, save human lives, and prevent the formation of new traffic jams and accidents.

CNews Analytics: Atlas of the Russian Internet of Things Market

Internet of Things

“We are constantly training the neural network on the reactions of car owners to an accident, many logics of human behavior are already included. For example, if a person began to get out of the car, go around it, put a red triangle, or stand for a long time, then an accident is very likely to happen. The module “sees” these behavioral patterns and gives a signal. The operator will react faster to an accident, which means that the time to rescue those involved in an accident will be longer. This is important when minutes count, nobody canceled the “golden hour” rule in medicine,” said the CEO of TrafficData Ivan Chebykin.

Source link

Leave a Reply

Your email address will not be published.