Floorplan-SLAM: A Real-Time, High-Accuracy, and Long-Term Multi-Session Point-Plane SLAM for Efficient Floorplan Reconstruction

TL;DR

Floorplan-SLAM presents a real-time framework that tightly integrates floorplan reconstruction with a stereo point-plane SLAM system on CPU, enabling long-term, multi-session indoor floorplan reconstruction using only stereo cameras. It introduces a robust plane extraction pipeline in a compact Plane Parameter Space, and a floorplan reconstruction module that formulates a binary linear programming problem over candidate wall segments under trajectory constraints, fused with continuous SLAM optimization and map merging for consistency. Across VECtor and self-collected datasets, the approach outperforms state-of-the-art methods in plane extraction robustness, pose estimation accuracy, and floorplan fidelity, achieving 25–45 FPS without GPU and reconstructing 1000 m^2 in about 9.4 minutes versus offline methods taking hours. This real-time, cost-effective solution enables scalable, long-term indoor mapping and provides valuable geometric priors for navigation and scene understanding in robotics.

Abstract

Floorplan reconstruction provides structural priors essential for reliable indoor robot navigation and high-level scene understanding. However, existing approaches either require time-consuming offline processing with a complete map, or rely on expensive sensors and substantial computational resources. To address the problems, we propose Floorplan-SLAM, which incorporates floorplan reconstruction tightly into a multi-session SLAM system by seamlessly interacting with plane extraction, pose estimation, and back-end optimization, achieving real-time, high-accuracy, and long-term floorplan reconstruction using only a stereo camera. Specifically, we present a robust plane extraction algorithm that operates in a compact plane parameter space and leverages spatially complementary features to accurately detect planar structures, even in weakly textured scenes. Furthermore, we propose a floorplan reconstruction module tightly coupled with the SLAM system, which uses continuously optimized plane landmarks and poses to formulate and solve a novel optimization problem, thereby enabling real-time incremental floorplan reconstruction. Note that by leveraging the map merging capability of multi-session SLAM, our method supports long-term floorplan reconstruction across multiple sessions without redundant data collection. Experiments on the VECtor and the self-collected datasets indicate that Floorplan-SLAM significantly outperforms state-of-the-art methods in terms of plane extraction robustness, pose estimation accuracy, and floorplan reconstruction fidelity and speed, achieving real-time performance at 25-45 FPS without GPU acceleration, which reduces the floorplan reconstruction time for a 1000 square meters scene from over 10 hours to just 9.44 minutes.

Figures (7)

• Figure 1: Floorplan reconstruction results on the meeting rooms sequence from the self-collected dataset by Floorplan-SLAM. Floorplan-SLAM can robustly process stereo images containing low-texture and cluttered objects at 45 FPS on a CPU, and it merges multiple sub-sessions recorded at different times, ultimately achieving real-time, high-accuracy, and long-term floorplan reconstruction.
• Figure 2: System Overview. Floorplan-SLAM tightly integrates a stereo point-plane SLAM system with a floorplan reconstruction module. The former provides accurate plane landmarks and trajectory information, while the latter formulates and solves a novel optimization problem, ultimately achieving real-time, high-accuracy, and long-term floorplan reconstruction.
• Figure 3: Comparison of floorplan reconstruction results with and without trajectory constraints. The top-left shows the CAD floor plans provided by the meeting room scenario in the self-collected dataset. The red box indicates the reconstruction results without trajectory constraints, while the green box represents the reconstruction results with trajectory constraints. The trajectory constraints ensure that the reconstructed floorplan contains navigable openings, thereby significantly improving reconstruction accuracy.
• Figure 4: Comparison of plane extraction results on the VECtor (upper) and self-collected (lower) datasets. ORB features are represented by crosses, and Sobel features are represented by dots in our method. The plane boundaries in both RSS and our approach are determined by the convex hull of the corresponding plane points.
• Figure 5: Our self-collected dataset setup.