CASPIAN JOURNAL
MANAGEMENT AND HIGH TECHNOLOGIES
METHOD OF BOUNDARIES SEQUENTIAL RР•FINING IN THE IMAGES OF TRAFFIC CONDITIONS
 |
Read | Lyasheva Stella A., Shleymovich Mikhail P. METHOD OF BOUNDARIES SEQUENTIAL RР•FINING IN THE IMAGES OF TRAFFIC CONDITIONS // Caspian journal : management and high technologies. — 2020. — №4. — pp. 21-31. |
Lyasheva Stella A. - Kazan National Research Technical University named after A.N. Tupolev-KAI, salyasheva@kai.ru
Shleymovich Mikhail P. - Kazan National Research Technical University named after A.N. Tupolev-KAI, mpshleymovich@kai.ru
Currently, the concept of "smart city" is being actively developed, which defines the tasks of optimal resource allocation and ensuring the safety of urban infrastructure, including road safety. This is especially true in connection with the growing number of vehicles, among which the appearance of self-driving cars is predicted. The solution to this problem is based on the use of various technologies, including computer vision technologies. One of the directions here is based on machine learning, in which models of objects in images are considered as feature vectors. Such models are descriptions of objects for training classifiers that ensure their detection and recognition. Boundary features are often used as attributes, which are formed by performing two steps: selecting boundaries and describing them as descriptors. This paper describes a method for selecting the boundaries of objects in images of traffic conditions, which is based on the use of multiple-scale wavelet transform. The method is based on determining the significance of the brightness change at a certain point at a certain level of the wavelet decomposition based on the estimation of the contribution to the total image energy of the corresponding detail coefficients. The method defines a sequential refinement of boundaries, which is done under the assumption that the boundary points match at different levels. This assumption is based on the fact that the brightness values of pixels of image copies at different levels of multiple-scale decomposition are interrelated with each other. The described method is simple to implement, has a relatively high speed and can be flexibly configured for different operating conditions.
Key words: self-driving cars intelligent systems, computer vision, detection and recognition of objects in images, object boundary features, boundaries detection on images, wavelet transform, image energy, boundary thinning