Abstract

Augmented Reality applications benefit from the use of SLAM systems, that can provide low latency and precise camera poses and 3D reconstructions of the environment. On the other hand, these methods lack semantic knowledge, which would enable the AR applications to adapt to the surroundings and to interact with it. The thesis has the goal of integrating a neural network for semantic segmentation into the ORB-SLAM2 system in order to enrich its 3D map with this additional knowledge. It also focuses on the research and the training of efficient network architectures and on how to use the GPS coordinates of the user position to enhance the models’ prediction accuracy.

Results/Implementation/Project Description

Conclusion

[ PDF (optional) ] 

[ Slides Kickoff/Final (optional)]