Application of the Weight Model to Detect the State Registration Number of the Vehicle in the Image

The article deals with the issues of detecting state registration numbers of vehicles in images when solving the problem of their identification in intelligent transport systems. The main approaches for solving this problem are described, which belong to two classes–based on image processing and based on machine learning. It is shown that, despite the effectiveness of methods based on machine learning, it is advisable to use methods based on image processing for certain specific conditions. Among the image processing methods for detecting vehicle license plates, edge analysis methods are the fastest. At the same time, you can use the properties of the wavelet transform to analyze the edges. This is the basis of the approach proposed by the authors, which provides for the use of a weight model based on the coefficients of the multiple-scale orthogonal Haar wavelet transform. As a result of building such a model, weights are calculated that reflect the significance of the pixels of the original image for the analysis of the edges when searching for the area of the vehicle number. The weight model allows you to emphasize the characteristics of the vehicle number area in the image and use additional processing to find its location. The advantages of the proposed approach include simplicity in implementation, the detection of vehicle numbers in images obtained under various conditions, and the possibility of supplementing it with operations that ensure increased efficiency at all stages.