| Abstract [eng] |
The purpose of the final master thesis is to facilitate engineer work taking into account scientific literature on similar form object search systems and create an easy-to-use prototype of application that can recognize shape parts from photos and provide object class probability. The thesis involves an investigation of existing 3D object search systems and their use. The machine learning and 3D object recognition principles are also investigated. After literature analysis, it is concluded that program prototype should be created using multi-view 3d object recognition together with Google Colaboratory environment and convolutional neural networks. UML diagrams represent the idea of realization. Also, the assessment of Google Colaboratory and 2D image recognition is made. Program prototype is capable of recognizing object class and can return a result if the object is screw or profile. Additionally, the quality of learning model and deficiencies of such a program prototype are evaluated. Conclusion and recommendations are presented. The thesis consists of 60 pages of text without appendixes, 25 pictures, 3 tables, 60 bibliographical entries. Appendixes are included separately. |
