Software (SW) that automatically recognizes road infrastructure objects and analyzes traffic has been developed at Novosibirsk State Technical University (NSTU). It does not require high computing power to operate, the university's press service reported.
The program is based on an algorithm using a convolutional neural network (a class of deep networks used for processing images, video, audio), said the project's scientific director, Egor Antonyants.
It detects markings, signs, traffic lights, pedestrian crossings, vehicles. Then the algorithm analyzes their location and generates a text description of the scene.
As a result, the program draws conclusions about traffic: low/medium road congestion or high traffic intensity, Antonyants explained.
Unlike simple algorithms that only identify objects on the road, NSTU's development analyzes video in real time.
The project authors believe that this tool can be useful for road services and insurers. In the future, it can be integrated into urban video surveillance systems, the university concluded.
Read more on the topic:
- A neural network has been created in the Urals that predicts traffic jams in advance using street cameras
- "Smart" traffic lights against traffic jams: Rostec tests adaptive DS7-Light in Nizhny Tagil
- Traffic jams will be calculated by braking: PNRPU scientists have created a program to predict traffic jams on roads before they appear