NDNakju Lett Doh

Nakju Lett Doh

*This is an algorithmically generated author record based on publicly available information.

Nakju Lett Doh has a total of 66 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Practical Framework for Visual Positioning in Real-World Indoor Environments

    by Bumchul Jang, Hyunga Choi, Nakju Doh, Janghun Hyeon

    A visual positioning system determines a user’s six degrees-of-freedom pose by analyzing images captured by cameras. This is crucial for applications in robotics, augmented reality, and indoor navigation. Despite advancements in such technology, creating datasets and frameworks for frequently changing indoor spaces within industrial settings remains challenging. Conventional datasets are small-scale, slow to collect, and rely on inefficient stationary devices, making them unsuitable for such environments. The proposed framework is designed to address these shortcomings and for practical application. It utilizes a LiDAR-based mobile mapping system for efficient data handling and employs a multiple-image query approach for robust localization. Additionally, the framework uses cluster-wise pose integration for multiple-image queries to address data sparsity and similarity problems in indoor localization. The results from experiments in diverse and challenging indoor spaces demonstrate that, even with extensive queries, the proposed method outperforms existing methods in localization accuracy, achieving robust performance with 96.5% of the estimated poses within a 1.0m error of the ground truth and an average error of 0.779m.

  • CLoc: Confident Initial Estimation of Long-Term Visual Localization Using a Few Sequential Images in Large-Scale Spaces

    by Joohyung Kim, Janghun Hyeon, Hyunga Choi, Bumchul Jang, Bokyeon Jeong, Nakju Doh

    Estimation of six degree-of-freedom camera pose with images is a fundamental task for various applications, such as autonomous driving and augmented reality. The challenges in visual localization stem primarily from the scale of spaces and the changes that can happen during the long-term intervals between when the database and query images are captured. Although most sequence-based visual localization (SVL) methods rely on the outputs of one-shot visual localization (OVL) for the initial estimation of the camera pose, OVL may yield incorrect poses because of these limitations. In this study, we aim to achieve a confident initial estimation using a few sequential images. Accordingly, we propose a pipeline called CLoc that narrows down possible solution spaces with hypotheses generated by associating pose candidates of keyframes and their relative poses to determine the most probable pose of the latest keyframe. Moreover, we create publicly available visual localization datasets including three building-scale indoor spaces. The datasets provide long-term challenges and sequential query images simultaneously, enabling the evaluation of the OVL and SVL methods in large-scale spaces. We evaluate CLoc using our datasets and a city-scale outdoor dataset and compare it with two state-of-the-art OVL methods and three SVL methods. By properly leveraging sequential data, CLoc can outperform other methods and achieves 100% accuracy in the [2m, 10°] evaluation thresholds for all datasets.


  • Open Access

    Robust Cuboid Modeling from Noisy and Incomplete 3D Point Clouds Using Gaussian Mixture Model

    by Woonhyung Jung, Janghun Hyeon, Nakju Doh

    A cuboid is a geometric primitive characterized by six planes with spatial constraints, such as orthogonality and parallelism. These characteristics uniquely define a cuboid. Therefore, previous modeling schemes have used these characteristics as hard constraints, which narrowed the solution space for estimating the parameters of a cuboid. However, under high noise and occlusion conditions, a narrowed solution space may contain only false or no solutions, which is called an over-constraint. In this paper, we propose a robust cuboid modeling method for point clouds under high noise and occlusion conditions. The proposed method estimates the parameters of a cuboid using soft constraints, which, unlike hard constraints, do not limit the solution space. For this purpose, a cuboid is represented as a Gaussian mixture model (GMM). The point distribution of each cuboid surface owing to noise is assumed to be a Gaussian model. Because each Gaussian model is a face of a cuboid, the GMM shares the cuboid parameters and satisfies the spatial constraints, regardless of the occlusion. To avoid an over-constraint in the optimization, only soft constraints are employed, which is the expectation of the GMM. Subsequently, the soft constraints are maximized using analytic partial derivatives. The proposed method was evaluated using both synthetic and real data. The synthetic data were hierarchically designed to test the performance under various noise and occlusion conditions. Subsequently, we used real data, which are more dynamic than synthetic data and may not follow the Gaussian assumption. The real data are acquired by light detection and ranging-based simultaneous localization and mapping with actual boxes arbitrarily located in an indoor space. The experimental results indicated that the proposed method outperforms a previous cuboid modeling method in terms of robustness.


  • Precise Camera–LiDAR Extrinsic Calibration Based on a Weighting Strategy Using Analytic Plane Covariances

    by Gunhee Koo, Jaehyeon Kang, Bumchul Jang, Nakju Doh

    Planarity is a practical feature for estimating the camera–LiDAR extrinsic parameters. LiDAR directly measures the 3-D plane in the form of a point cloud, but the camera captures the 2-D plane image which requires perspective-n-point (PnP) techniques for 3-D plane estimation. Likewise, these 3-D plane acquisition processes by LiDAR and camera are quite different; hence, their plane covariances from each sensor also differ. The plane measurement covariance originates from the LiDAR measurement noise; it can be directly calculated from a 3-D point cloud. The plane parameter covariances originate from the corner point errors in a 2-D image; however, their direct formulas have not been studied. In this study, we analytically derive the plane parameter covariances under the assumption of linear error propagation. To validate the proposed covariances, we implemented a simulation to observe how effectively the proposed covariances reflect on the true plane parameter covariances. Next, we validated that the proposed covariances could increase the performance of the camera–LiDAR extrinsic calibration. We compared our method with other calibration methods by varying the pixel noise levels and the number of board measurements in calibration simulations. In calibration field tests, we calibrated a scanning system equipped with a Ladybug5+ camera and a VLP-16 LiDAR and compared our method with other calibration methods.


  • Reduction of LiDAR Point Cloud Maps for Localization of Resource-Constrained Robotic Systems

    by Chungjae Choe, Sangmin Ahn, Nakju Doh, Changjoo Nam

    This article presents a method for mobile robotic systems to reduce the size of 3-D point cloud maps while achieving small localization errors even though the reduced maps are used for localization. Localization with dense depth maps especially in large environments could be computationally expensive, so not affordable for robotic systems with limited computational resources. Since such point cloud data collected using a multichannel 3-D LiDAR have representational redundancy, some amount of the data could be removed for compactness. However, existing methods demand annotations and descriptors defined by human experts or lose important spatial features for accurate localization. We propose a fully-automatic reduction method for reliable navigation of mobile systems with low computing power and limited storage. Given a full 3-D point cloud map, our method generates a graph whose vertices and edges represent robot poses and the similarity between the data collected from a pair of robot poses, respectively. Each robot pose is associated with a point cloud set. We find a dominating set of the graph, which is a subset of robot poses such that the robot can localize reliably only with the point clouds collected from those poses. Experimental results with indoor environments show that our system reduces the amount of map data up to 93% while not exceeding acceptable ranges of average translation and rotation errors, which are 0.25 m and 0.5 $^\circ$ , respectively.


  • A Two-Step Optimization for Extrinsic Calibration of Multiple Camera System (MCS) Using Depth-Weighted Normalized Points

    by Gunhee Koo, Woonhyung Jung, Nakju Doh

    A multiple camera system (MCS), which allows only a limited common field-of-view between adjacent cameras, has been chiefly calibrated using a typical 2D-3D point correspondence. However, the correspondence contains a potential instability by which the calibration result can diverge, and the instability has not been studied nor overcome before. We propose a MCS extrinsic calibration method with high robustness and accuracy based on a two-step optimization strategy. Using depth-weighted normalized points, we develop two novel types of point correspondence as follows. The 1st correspondence, F cb , aims to robustly estimate the MCS extrinsic parameters by overcoming the potential instability that exists in the typical 2D-3D point correspondence. The 2nd correspondence, F cc , aims to refine the MCS extrinsic parameters by using the direct relation between adjacent cameras. In the simulation, we validated the robustness and high accuracy of the proposed method. We validated its high precision in the field test. In both the simulation and the field test, our method was compared with the state-of-the-art method.


  • Open Access

    Automatic Reconstruction of Multi-Level Indoor Spaces from Point Cloud and Trajectory

    by Gahyeon Lim, Nakju Doh

    Remarkable progress in the development of modeling methods for indoor spaces has been made in recent years with a focus on the reconstruction of complex environments, such as multi-room and multi-level buildings. Existing methods represent indoor structure models as a combination of several sub-spaces, which are constructed by room segmentation or horizontal slicing approach that divide the multi-room or multi-level building environments into several segments. In this study, we propose an automatic reconstruction method of multi-level indoor spaces with unique models, including inter-room and inter-floor connections from point cloud and trajectory. We construct structural points from registered point cloud and extract piece-wise planar segments from the structural points. Then, a three-dimensional space decomposition is conducted and water-tight meshes are generated with energy minimization using graph cut algorithm. The data term of the energy function is expressed as a difference in visibility between each decomposed space and trajectory. The proposed method allows modeling of indoor spaces in complex environments, such as multi-room, room-less, and multi-level buildings. The performance of the proposed approach is evaluated for seven indoor space datasets.