Detecting 3D Line Segments for 6DoF Pose Estimation with Limited Data

TL;DR

This work tackles 6DoF pose estimation for industrial bins under limited real data by exploiting the cuboid geometry of bins. It introduces a 3D extension of the LeTR line-segment detector to structured point clouds and uses four top-edge line segments as geometric primitives to recover the bin pose, avoiding instance-specific CAD models. The authors also release a new dataset combining real and synthetic scans and show that synthetic data substantially boosts real-scan accuracy. Empirically, the approach achieves notable improvements over state-of-the-art methods in translation and rotation accuracy (approximately 3 cm and 8°) and demonstrates practical viability for industrial bin handling. The work further provides ablations, a detailed training regimen, and discusses limitations and future directions for robust line-based pose estimation.

Abstract

The task of 6DoF object pose estimation is one of the fundamental problems of 3D vision with many practical applications such as industrial automation. Traditional deep learning approaches for this task often require extensive training data or CAD models, limiting their application in real-world industrial settings where data is scarce and object instances vary. We propose a novel method for 6DoF pose estimation focused specifically on bins used in industrial settings. We exploit the cuboid geometry of bins by first detecting intermediate 3D line segments corresponding to their top edges. Our approach extends the 2D line segment detection network LeTR to operate on structured point cloud data. The detected 3D line segments are then processed using a simple geometric procedure to robustly determine the bin's 6DoF pose. To evaluate our method, we extend an existing dataset with a newly collected and annotated dataset, which we make publicly available. We show that incorporating synthetic training data significantly improves pose estimation accuracy on real scans. Moreover, we show that our method significantly outperforms current state-of-the-art 6DoF pose estimation methods in terms of the pose accuracy (3 cm translation error, 8.2$^\circ$ rotation error) while not requiring instance-specific CAD models during inference.

Figures (4)

• Figure 1: Our 6DoF bin pose estimation pipeline from structured point cloud. To estimate the 6DoF pose we first detect 4 3D line segments. A robust geometric regression scheme is then used to obtain the final pose.
• Figure 2: Top: 3D line segments detected by our method and the annotated line segments. Bottom: Estimated pose based on the predicted line segments compared with the annotated pose.
• Figure 3: Samples from the extended dataset of 3D scans of bins. Green lines indicate the annotated bin positions. The three images on the left show real scans, while the image on the right shows synthetically generated scans.
• Figure 4: Lines predicted by our best model on samples from the test set. The leftmost image depicts a bin scan with a missing corner; despite this, the model is able to infer the position of the missing corner.