Neural Network in Smartphone Helps Visually Impaired People Detect Potholes and Stairs
Novosibirsk entrepreneurs have developed Videonavigator, a system for visually impaired people that uses images from a smartphone camera to detect obstacles and determine how far away they are.
The app uses computer vision and neural network algorithms to analyze images in real time. It recognizes more than 30 classes of objects, including potholes and stairs, and detects obstacles up to 5 meters away. The app works offline, with a latency of less than 250 milliseconds.
The developers tested the system on Android with the participation of pilot users and experts from the All-Russian Society of the Blind. Field trials were conducted in accordance with GOST requirements and confirmed the global novelty of wavelet analysis in assistive technologies. Detection accuracy for hazardous objects exceeded 80%.
The system has been integrated into the IDEYSTVIE platform and is expected to be introduced in the Novosibirsk Region, other Russian regions and the markets of the Eurasian Economic Union.
