Power-efficient Visual Perception for AR Glasses

In this project, we try to create a power-efficient visual perception system that seeks to save on-device power while not compromising SLAM (simultaneous localization and mapping) trajectory outputs.

Weekly progress

Week 1

• Manually decreased framerate of file and then saw MPS error
• Edited metadata tags (specifically nominal_rate) using VRS library, got MPS working

Week 2

• Ran MPS on many sequences and then put results in Google Sheets
• Compared multi-sequence results for different framerates

​Presentation

Week 3

• Started to investigate the ego_blur repo and obtained rectified images, transformation jsons

Other info

• This is the command to compile files that use the VRS library:

      g++ FILE_NAME.cpp -o EXECUTABLE_NAME \
          -L/usr/local/lib \
          -I/usr/local/include \
          # VRS Libraries (from most dependent to least, or internal dependencies last)
          -lvrs_utils \
          -lvrs_utils_converters \
          -lvrs_utils_cli \
          -lvrs_helpers \
          # Libraries that define functions used by the above, and also by vrslib
          -lvrslib \
          -lvrs_os \
          -lvrs_logging \
          -lvrs_utils_xxhash \
          # External Libraries (order generally less critical, but still good to put providers later)
          -llz4 \
          -lzstd \
          -lboost_filesystem \
          -lboost_system \
          -lxxhash \
          -lfmt \
          -lpthread