Comparison of Analyze-Then-Compress Methods in Edge-Assisted Visual SLAM