MDF: Multi-Modal Data Fusion with CNN-Based Object Detection for Enhanced Indoor Localization Using LiDAR-SLAM

TL;DR

The paper tackles indoor localization in GPS-denied environments by proposing a cost-effective handheld system that fuses 2D LiDAR, an IMU, and CNN-based object detection within a Cartographer ROS framework. The approach integrates CNN-derived object landmarks into a tightly coupled LiDAR-IMU fusion with IEKF, followed by object-assisted pose and map optimization and global pose-graph refinement, achieving notable improvements in localization performance. Key contributions include a feature-based LiDAR separation into edge and surface points, a Madgwick+EKF IMU denoising pipeline, an IEKF-based LiDAR-IMU fusion, a CNN model that outputs up to four object landmarks, and landmark-assisted SLAM with dynamic-object management and global optimization. Experimental results demonstrate significant reductions in ATE (around $21.03\%$) and RMSE improvements, along with high object-detection accuracy ($\text{Precision}=92.5\%$, $\text{Recall}=91.3\%$, $\text{F1}=0.919$) on real indoor corridors and KITTI-like datasets, all while maintaining real-time performance (~$15$ FPS on CPU) and avoiding loop closures. The work shows strong potential for robust, scalable indoor localization in robotics, emergency response, and industrial automation, with clear directions for extending to 3D sensing and memory-efficient deployment.

Abstract

Indoor localization faces persistent challenges in achieving high accuracy, particularly in GPS-deprived environments. This study unveils a cutting-edge handheld indoor localization system that integrates 2D LiDAR and IMU sensors, delivering enhanced high-velocity precision mapping, computational efficiency, and real-time adaptability. Unlike 3D LiDAR systems, it excels with rapid processing, low-cost scalability, and robust performance, setting new standards for emergency response, autonomous navigation, and industrial automation. Enhanced with a CNN-driven object detection framework and optimized through Cartographer SLAM (simultaneous localization and mapping ) in ROS, the system significantly reduces Absolute Trajectory Error (ATE) by 21.03%, achieving exceptional precision compared to state-of-the-art approaches like SC-ALOAM, with a mean x-position error of -0.884 meters (1.976 meters). The integration of CNN-based object detection ensures robustness in mapping and localization, even in cluttered or dynamic environments, outperforming existing methods by 26.09%. These advancements establish the system as a reliable, scalable solution for high-precision localization in challenging indoor scenarios

Figures (9)

• Figure 1: CNN-LiDAR-SLAM system overview.
• Figure 2: Proposed CNN model architecture. The network takes as input a 26-dimensional feature vector derived from LiDAR and IMU fusion and outputs 12 object parameters: center positions $(x, y)$ and radii for up to four detected objects.
• Figure 3: Performance comparison of object detection between the proposed method (in navy) and bdcc7010043 (in dark red). The four shapes represent items of varying sizes, classified by radius into four categories. Dashed ovals highlight regions where the proposed method achieves improved object detection accuracy compared to bdcc7010043.
• Figure 4: Overview of the experimental corridors, showing 2D maps with trajectories and marked features.
• Figure 5: CPU and memory usage comparison for CNN-LiDAR-SLAM, A-LOAM, and SC-ALOAM across two datasets: KITTI Dataset 07 and Indoor Dataset.