Moving objects recognition method by their video images
Михаил Александрович Анохин · Eastern-European Journal of Enterprise Technologies · 2014
Modern computer systems have sufficient capacity to significantly improve the efficiency of the image recognition system. These opportunities allow to prevent man-made disasters and social conflicts, so this area of research is in demand in the contemporary community. Systems in most cases have a modular structure, and feature-formation and decision-making modules are one of the most important components. A brief analysis of existing methods was given in the paper, and promising directions for developing own solutions of the problem were selected. Feature-formation method is based on using the orthogonal transformation of the spatial spectrum of the video image, which allows to eliminate the effect of some deformations on the recognition quality, as well as to reduce the set of features, used in the decision-making. Since the set of features is represented by the eigenvectors, the decision rule construction algorithm is based on using the Dice similarity criterion. This criterion was chosen because it has allowed to more qualitatively compare the feature vectors of the input image and the image from the database. The results of performance evaluation of the recognition system prototype were given. In the future, using large data arrays in order to optimize the system and increase the evaluation reliability of its quality characteristics is expected.