Approach Using Color Images and Machine Learning
The object recognition from 2D images based on learning is focused on exploring the classical machine learning and deep learning approaches for object detection and recognition using 2D images. The goal of these methods is quickly determined the object types and locations in the images. To train these kinds of recognizers, the sufficient amount of training data is needed, therefore, our training data are mainly obtained from the synthetic 3D model. Obtained objects information (type, location) can be used as the first step of the pose detectors that are developed in the next parts of this project. At this moment, we are testing the multiple-view object detector that is capable to detect different objects in the real time using convolutional neural networks (CNN), however, the performance of this recognition approach is negatively influenced by the complex background. In the next steps, we will experiment with more complex architectures of CNN that can achieve better results.