Detekcia obalových kvádrov okolo objektov v obraze

Abstract

The main theme of this work is the detection of 3D bounding box around objects in image. At the beginning, we had to do a survey to find out which methods are currently being used and what results they are achieving. We chose the DENSEBOX method because it is relatively fast and the output is very easy to edit. The object detection is done by convolution neural network. During designing NN we were inspired by the work [18] in which NN detects objects at different scales. We used the Tensorflow framework, for which the NN implementation had to be modified but the architecture remained the same. It is important to select the dataset appropriately for the selected method. We decided to use a KITTI dataset that is sufficiently comprehensive and well described. The result of this work is program which uses neural network to detect bounding boxes various sizes and visualize the results.