RGB-D SLAM in Indoor Planar Environments With Multiple Large Dynamic Objects

Abstract

This work presents a novel dense RGB-D SLAM approach for dynamic planar environments that enables simultaneous multi-object tracking, camera localisation and background reconstruction. Previous dynamic SLAM methods either rely on semantic segmentation to directly detect dynamic objects; or assume that dynamic objects occupy a smaller proportion of the camera view than the static background and can, therefore, be removed as outliers. With the aid of camera motion prior, our approach enables dense SLAM when the camera view is largely occluded by multiple dynamic objects. The dynamic planar objects are separated by their different rigid motions and tracked independently. The remaining dynamic non-planar areas are removed as outliers and not mapped into the background. The evaluation demonstrates that our approach outperforms the state-of-the-art methods in terms of localisation, mapping, dynamic segmentation and object tracking. We also demonstrate its robustness to large drift in the camera motion prior.

RGB-D-Inertial SLAM in Indoor Dynamic Environments with Long-term Large Occlusion

“…In the scenario of LTLO (seq. [6][7][8][9], evaluation demonstrates that our method is able to provide accurate camera trajectories and outperforms all other methods (Figure 5). While neither ORB-SLAM3 nor Dynamic-VINS is able track camera trajectory correctly.…”

“…against ground truth camera trajectories (Table II). The results are compared with SF [3], PF [7], VINS-Mono [16], ORB-SLAM3 [9] and Dynamic-VINS [19]. In static environments, our method achieves comparable results with other state-of-the-art visual-inertial SLAM methods.…”

“…The novelty of formulation 1 is that we actively estimate dense segmentation of input images Γ and sparse segmentation of map points using a combination of dense and sparse features. This is different to StaticFusion [3] or PlanarFusion [7] which only consider the segmentation of images. We can, therefore, recover a static sparse map and correct IMU biases even if the initial dense segmentation of images are unreliable.…”

“…It also actively detects temporarily static objects to reject false loop closure constraints. Our previous work RigidFusion [6] uses camera motion prior from wheel odometry to simultaneously track the camera and a single rigid object that causes large occlusion, which is then extended to multiple planar object tracking [7]. However, none of these methods has demonstrated their performance if the large occlusion in the camera view lasts for the majority of time when the robot perceives the world.…”

“…Various dynamic SLAM methods [6], [7], [8] have explicitly considered dynamic objects that are dominant in the camera views with the aid of robot proprioception, like wheel odometry or Inertial Measurement Units (IMU). Some static visual-inertial navigation system (VINS) methods [9], [10] have also demonstrated their robustness when the camera view is fully occluded for a short period.…”