Revisiting Constraints for Monocular Visual Quadric SLAM

In recent years, numerous objectoriented SLAM systems have been proposed to enhance scene understanding and human-computer interaction by reconstructing semantic objects. Although significant progress has been made in semantic object reconstruction, these methods have not led to effective improvements in localization accuracy. In this paper, we systematically review the existing constraints used in visual quadric-based SLAM, revealing their limitations-particularly in handling mismatches between complex object contours and dual quadrics. To quantitatively evaluate the performance of these constraints, we conduct experiments on public datasets, which demonstrate the impact of different constraints on dual quadric SLAM methods and validate our analysis.