Vasily E. Gai, Alexander V. Smirnov, Roman O. Barinov, Igor V. Polyakov, Vladimir A. Golubenko, Georgy D. Kuznetsov

Nizhny Novgorod State Technical University

In this paper we propose a method for detecting an object in an image based on a global feature description. An information model is described and implementation options for each of the stages of the image object detection task are proposed. For the image preprocessing stage, the available options, including normalization, calculation of brightness function and application of Gaussian filter are given. The stage of forming a global feature description of an object is based on active perception theory (U-transformation). Object localization is performed based on the minimum Euclidean distance from the detected object to the reference objects from the database. Images from the Russian Traffic Sign Dataset database and their modified copies (images with superimposed noise, images with rotation of the objects searched for) were used for testing. When analyzing the test results, the parameters that give the highest accuracy for the proposed method of object detection in the image have been selected. In the presence of noise in the image, the localization accuracy of the proposed method was more than 70%. The proposed image object detection method is robust to rotation of the objects being searched. The resulting accuracy of about 94-96%, when compared with the accuracy of existing methods, showed that under normal conditions the developed method works similarly to existing methods.

image processing, object detection, theory of active perception