P2U-SLAM: A Monocular Wide-FoV SLAM System Based on Point Uncertainty and Pose Uncertainty [PDF]

Yufan Zhang, Kailun Yang, Ze Wang, Kaiwei Wang

Video (speed ×3)

P2U-SLAM run on TUM-VI Benchmark

The sequence Corridor4 includes a scenario of rapid back-and-forth movement in a corridor, which poses a moderate challenge to the accumulated error and scale consistency of SLAM systems.

While the sequence Magistrale3 corresponds to a large - scale indoor loop trajectory. What's quite interesting is that, thanks to the suppression of the impact of weak constraints on optimization by point uncertainty and pose uncertainty, P2U-SLAM has a very small cumulative error in this 566-meter-long sequence. When the trajectory loops, there is no need for the intervention of the loop module, and the co-visibility constraints of the old keyframes can be automatically identified only through local map search (from 00:36 to 00:40 in the video). However, all the SLAM algorithms we compared in the experimental part of the paper triggered loop closure correction on this sequence. Even so, the fianl accuracy of P2U-SLAM on the Magistrale3 sequence is still the highest among all the comparative methods.

Corridor4	Magistrale3
P2U-SLAM--TUM-VI-Corridor4.mp4	P2U-SLAM--TUM-VI-Magistrale3.mp4

P2U-SLAM run on PALVIO-EX Dataset IDL01

This is the recording result of our team's test in a large-scale indoor scene, with a total distance exceeding 1km. The starting point and endpoint of the route coincide, and at the end of the video, it can be seen that the initial keyframe (red&thic) perfectly overlaps with the final frame (green&thic) of the camera. (The loop module was not triggered throughout the entire tracking process)

P2U-SLAM--PALVIO-IDL01.mp4

Method

Illustrations of situations in which point and pose uncertainty are applied. (a) The map points are viewed as fixed when estimating the pose of a new frame of tracking, so the point uncertainty should be applied according to Eq. (12) and Eq. (14) in P2U-SLAM. (b). In the process of local BA, there are some fixed keyframes viewed as observation results that are not targets to estimate while still having a great influence on optimization. Similarly, the pose uncertainty of these fixed keyframes (with orange dotted boxes) should be processed as Eq. (17) and Eq. (18) in P2U-SLAM.

Pipline

The pipeline of P2U-SLAM. P2U-SLAM mainly consists of initialization, tracking, local BA, and loop closing. Point uncertainty functions in the tracking module to suppress the noise from treating past map points as measurement results on the current frame¡¯s pose estimation. Pose uncertainty acts on the local BA module to suppress the noise from treating fixed keyframe poses as measurement results on other variables to estimate.

Experiments

P2U-SLAM is evaluated on two open-source wide-FoV SLAM dataset. The ATE results of serveral algorithms on PALVIO dataset. Each algorithm runs a total of ten times on each sequence, with the results of each run represented by the color squares. A higher red component in the square indicates a worse result in the corresponding metric test, while a greater blue component signifies better performance. An incomplete run is represented by a blank white square. Stability comparison on sequences from TUM-VI dataset. Each node on the line chart corresponds to a sequence in the dataset.