Specialists from the Kazan National Research Technical University named after A. N. Tupolev-KAI have created a neural network for unmanned vehicles. This computer vision system recognizes the colors of traffic light signals with an accuracy of over 90%, even in bad weather. More than 40 thousand images were used to train the neural network, including traffic lights in different weather conditions.
Our system not only determines the color of the traffic light signal, but also accurately localizes it in the image. This is critical to ensuring the safety of unmanned vehicles in urban environments.
A key feature of the development is the ability to adapt to different lighting conditions. The neural network can recognize near and far traffic lights, as well as simultaneously track multiple signals.
The system has no problems recognizing red and green traffic light signals. The yellow signal is determined with less accuracy. According to scientists, the neural network correctly determines signals in 90.5% of cases. It can process up to 55 frames per second in real time.
Scientists believe that the introduction of such a development in unmanned vehicles will significantly improve road safety.
Read more on the topic:
«Auto.ru» and YandexGPT: neural network shares key moments in the history of car ownership
Yandex unmanned vehicles have mastered a more human driving style
