Scientists from the University of Science and Technology MISIS, together with CITYLABS experts, developed algorithms for CCTV cameras that detect blurred and illuminated vehicle numbers. This was reported to socialbites.ca from MISIS.
One of the important tasks that arise in the analysis of traffic situations, including the conditions of technological roads, is the identification of a particular vehicle by the state registration plate. Often times, cars are not recognized correctly due to the high speed of the car, bright headlights, dustiness and insufficient camera capabilities.
To determine the degree of illumination of a plate, experts suggest using the analysis of the brightness histogram. The well-known yolo-v5 neural network is used to detect both vehicles and license plates.
“When training neural networks to identify cars and numbers, datasets were created taking into account time of day, seasonality, and weather. After the area of the characters in the image is determined, the area selected from the three-dimensional RGB color space is reduced to one-dimensional “gray”. After the histogram was calculated, the portion responsible for the “overexposure” was selected, so that 95.7% of the numbers were correctly classified as overexposure. To determine the degree of turbidity, a neural network with a unique architecture was built on a PC that provides 96.4% classification accuracy with a minimum processing time of 0.073 ms,” said Igor Temkin, Head of Automated Control Systems (ACS) at NUST MISIS.
A separate task during work on the neural network to determine blurring was the creation of a dataset for training. The developed algorithm provides a quantitative assessment of the degree of blur and illumination, as well as classifying images as readable and unreadable. This data can be used to adjust camera parameters such as shutter speed and aperture, which will improve the quality of later frames.
During the experiments, the effectiveness of the proposed approaches was demonstrated on various devices such as PC and Nvidia Jetson Nano microcomputer.
ancient scientists created flight delay estimation app.