C.DOT - Convolutional Deep Object Tracker for Augmented Reality Based Purely on Synthetic Data.

Area of Science:

• Computer Vision
• Machine Learning
• Augmented Reality

Background:

• Object tracking is crucial for augmented reality (AR) applications, enabling virtual content overlay.
• Current industrial object tracking systems often have complex manual configuration, hindering usability for service engineers.

Purpose of the Study:

• To investigate replacing manual configuration in object tracking with a machine learning approach.
• To develop an automated process for creating object tracker facilities using exclusively synthetic data.

Main Methods:

• An automated process for generating highly enhanced synthetic data was developed.
• A convolutional neural network was trained on this synthetic data for object tracking.
• The system was designed to work with simple RGB cameras for real-world applications.

Main Results:

• The automated synthetic data approach achieved superior performance compared to related work on the LINEMOD dataset.
• The method demonstrated reliable and robust results in real-world applications using RGB cameras.
• While performance for high-accuracy industrial demands is lower than manual methods, it offers significant initialization support.

Conclusions:

• Automated synthetic data generation offers a viable alternative to manual configuration for object tracking in AR.
• This machine learning-based approach enhances usability for service maintenance engineers.
• The system shows promise as a complementary tool for industrial AR applications, particularly during initialization phases.