Incomplete 3D motion trajectory segmentation and 2D-to-3D label transfer for dynamic scene analysis

Abstract

The knowledge of the static scene parts and the moving objects in a dynamic scene plays a vital role for scene modelling, understanding, and landmark-based robot navigation. The key information for these tasks lies on semantic labels of the scene parts and the motion trajectories of the dynamic objects. In this work, we propose a method that segments the 3D feature trajectories based on their motion behaviours, and assigns them semantic labels using 2D-to-3D label transfer. These feature trajectories are constructed by using the proposed trajectory recovery algorithm which takes the loss of feature tracking into account. We introduce a complete framework for static-map and dynamic objects' reconstruction, as well as semantic scene understanding for a calibrated and moving 2D-3D camera setup. Our motion segmentation approach is faster by two orders of magnitude, while performing better than the state-of-the-art 3D motion segmentation methods, and successfully handles the previously discarded incomplete trajectory scenarios.

Moving Object Detection by 3D Flow Field Analysis

“…This algorithm relies on the consistency of the tracked trajectories and is therefore sensitive to lost tracking situations and partial occlusions. To improve its robustness, [Jiang et al, 2017b] proposed a 3D-SMR algorithm which jointly benefits from the 2D-SMR in scalable feature size and tracking correspondence. In a more sophisticated manner, [Keuper et al, 2018] incorporate the low-level feature trajectory and high-level object recognition cues to achieve better performance.…”

Moving Objects Aware Sensor Mesh Fusion for Indoor Reconstruction From a Couple of 2d Lidar Scans

“…For robot applications, these moving objects should be detected to adapt the behaviour of the robot to its surrounding for scene mapping. For 3D reconstruction purposes, these objects should be removed in order for the 3D model to represent only the static part of the scene (Jiang et al, 2017a). In both cases, motion analysis is a mandatory preprocessing step.…”