The KITTI Vision Benchmark Suite

We develop a loop closing strategy that enhances
loop recall by integrating both geometric and
semantic consistency. To effectively utlize
semantic information, we propose a semantic
distance graph to organize semantic details. To
measure the similarity between semantic distance
graphs, we propose semantic object associations
after long interval. Finally, we implement an
object-level Bundle Adjustment (BA) method that
accurately computes geometric transformations
between matching keyframes, to improve loop recall
and mapping accuracy.

None

None