MineInsight: A Multi-sensor Dataset for Humanitarian Demining Robotics in Off-Road Environments

Mario Malizia, Charles Hamesse, Ken Hasselmann · 2025

Read MineInsight: A Multi-sensor Dataset for Humanitarian Demining Robotics in Off-Road Environments by Mario Malizia, Charles Hamesse, Ken Hasselmann (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2506-04842.txt.

Introduction

Mario Malizia and co-authors introduce a 2025 dataset for robots that search for landmines. Finding mines by hand is slow and risky, and robot vision needs good test data. The team placed 35 targets on three off-road tracks. The set held 15 mines and 20 common objects that can fool a search.

A ground robot and its arm scanned the tracks with many sensors. The set holds color, heat, and light-band images, plus laser scans, from day and night runs. Human-checked boxes mark each target. With about an hour of data, the set gives teams a shared test for mine-finding methods. It is free for others to use. Teams can compare results on shared ground.

Catalog record

CREATORMario Malizia, Charles Hamesse, Ken Hasselmann
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE30 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2506-04842

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

MineInsight: A Multi-sensor Dataset for Humanitarian Demining Robotics in Off-Road Environments Mario Malizia, Charles Hamesse, Ken Hasselmann Abstract The use of robotics in humanitarian demining increasingly involves computer vision techniques to improve landmine detection capabilities. However, in the absence of diverse and realistic datasets, the reliable validation of algorithms remains a challenge for the research community. In this paper, we introduce MineInsight, a publicly available multi-sensor, multi-spectral dataset designed for off-road landmine detection. The dataset features 35 different targets (15 landmines and 20 commonly found objects) distributed along three distinct tracks, providing a diverse and realistic testing environment. MineInsight is, to the best of our knowledge, the first dataset to integrate dual-view sensor scans from both an Unmanned Ground Vehicle and its robotic arm, offering multiple viewpoints to mitigate occlusions and improve spatial awareness. It features two LiDARs, as well as images captured at diverse spectral ranges, including visible (RGB, monochrome), visible short-wave infrared (VIS-SWIR), and long-wave infrared (LWIR). Additionally, the dataset comes with an estimation of the location of the targets, offering a benchmark for evaluating detection algorithms. We recorded approximately one hour of data in both daylight and nighttime conditions, resulting in around 38,000 RGB frames, 53,000 VIS-SWIR frames, and 108,000 LWIR frames. MineInsight serves as a benchmark for developing and evaluating landmine detection algorithms. Our dataset is available at https://github.com/mariomlz99/MineInsight . This work has been submitted to the IEEE for possible publication. Copyright may be transferred without notice, after which this version may no longer be accessible. I INTRODUCTION Landmines remain one of the most dangerous legacies of conflicts, posing long-term threats to civilians and obstructing post-war recovery efforts. In many affected regions, the presence of antipersonnel (AP) and antitank (AT) landmines severely limits agricultural activities, infrastructure development, and safe mobility. Traditional humanitarian demining methods, relying on human experience and metal detectors, are slow, hazardous, and costly. Consequently, robotic systems have been increasingly deployed [ 1 ] to accelerate landmine detection and improve operational safety. While many robotic systems rely solely on camera feedback for operators, researchers have increasingly explored diverse detection techniques [ 2 , 3 , 4 , 5 , 6 , 7 ] exploiting images captured across multiple spectral ranges. Existing landmine detection datasets (Table I ) often suffer from significant limitations, such as a narrow range of explosive ordnance types, limited sensor diversity, or inadequate coverage of challenging environmental conditions. These constraints on existing datasets hinder the development and validation of novel detection algorithms. Moreover, the lack of a standardized storage format for data releases makes it difficult to ensure consistency and comparability across studies, highlighting the need for datasets that follow standardized practices established in off-road mobile robotics datasets [ 8 , 9 , 10 ] . Most datasets rely on one sensor, neglecting sensor fusion applications incorporating other spectral ranges such as visible short-wave infrared (VIS-SWIR) or long-wave infrared (LWIR). Additionally, these datasets often omit LiDAR data, which delivers essential depth information for navigating uneven terrain, detecting obstacles, and improving landmine localization. They do not adequately address occlusion, a major challenge when mines are covered with vegetation, debris, or soil [ 11 ] . To the best of our knowledge, the majority of existing datasets in the literature are recorded using Unmanned Aerial Vehicles (UAVs). However, in many scenarios, deploying Unmanned Ground Vehicles (UGVs) is more practical or necessary, highlighting a gap that we aim to fill. To address these gaps, we introduce MineInsight , a multi-sensor, multi-spectral dataset designed to advance robotic landmine detection. MineInsight integrates dual sensor perspectives, capturing both 2D and 3D information from a UGV and its robotic arm, enabling a wider spatial understanding of landmines in complex terrains. Our dataset features two LiDARs and a diverse set of multi-spectral sensors spanning monochrome, RGB, VIS-SWIR, and LWIR spectra. The contributions of this paper are as follows: • We introduce MineInsight, a dataset featuring a diverse array of inert AP and AT landmines, along with common items that might be found in natural environments, to better mimic realistic demining scenarios and challenge detection systems against false positives. • Our dataset is, to the best of our knowledge, the first dataset that integrates dual sensor scans captured from both the UGV and its robotic arm. The robotic arm’s dynamic motion provides varying sensor viewpoints that effectively mitigate obstructions and enhance the detection of hidden landmines. • We provide an estimation of target locations, which include 3D positions on a global map and corresponding 2D bounding boxes in camera frames, serving as a benchmark for validating detection algorithms and facilitating further research and development in the field. II RELATED WORK Landmine detection datasets, as illustrated in Table I , are relatively rare in the existing literature. TABLE I: Comparison of different landmine detection datasets. Missing entries indicate unavailable data. Note: “AVision” is the short form for AlliedVision. Dataset Total Landmines Platform Environment Camera LiDAR Labels De Smet et al. [ 12 ] 1 UAV Grass, rubble Parrot Sequoia Multispectral – – Steinberg et al. [ 13 ] 1 UAV Grass, rubble Parrot Sequoia Multispectral – 2D bounding box Baur et al. [ 14 ] 1 UAV Sand Parrot Sequoia Multispectral – – Leloglu et al. [ 15 ] 1 Fixed Tripod Clay ATOM 1024 IR, FLIR T 650 SC – – Tamayo et al. [ 16 ] 1 UAV Low vegetation, clay Zenmuse XT IR – – Vivoli et al. [ 17 ] 2 Handheld Support Grass, gravel iPhone 13 – 2D bounding box MineInsight (Ours) 15 UGV Low to high vegetation, leaves AVision Alvium 1800 U-240C, AVision Alvium 1800 U-130, Sevensense Core Research, Teledyne FLIR Boson 640 Livox AVIA, Livox Mid-360 3D location and pixel coordinates of the center of each target This section offers an overview of the available ones, detailing their content and the conditions under which they were collected. Binghamton University has contributed significantly to the field of landmine detection by releasing multiple datasets as part of the Scatterable Landmine Detection Project. These datasets include data from various UAV flights flown over different terrains. The project’s primary focus is the detection of PFM-1 landmines, together with their carrying case. We propose a categorization of the dataset releases into three groups (1–7, 8–9, and 10–24), facilitating a direct comparison between UAV platforms, sensors, and data collection methods across studies. The first group (Datasets 1-7), released by De Smet et al. [ 12 ] , includes data collected using a DJI Phantom 4 Pro UAV and a DJI Matrice 600 Pro UAV, both equipped with a Parrot Sequoia Multispectral Camera. Flights were conducted over grass and rubble environments, resulting in seven high-resolution orthophotos. Steinberg et al. released the second group (Datasets 8-9) [ 13 ] , which includes data collected from three flights over grass and four flights over rubble using a DJI Matrice 600 Pro UAV equipped with a Parrot Sequoia Multispectral Camera. Baur et al. later used the same data to explore the potential of deep learning-based landmine detection, with their experimental setup well described in [ 3 ] , where they address the challenges of detecting PFM-1 landmines in UAV-based imagery. Baur et al. [ 14 ] released the third group (Datasets 10-24), which comprises

Read the complete text