Our Planetary Work

Super-3D Modelling

SAIIL's innovative Super-3D modelling technology achieves subpixel-scale topographic mapping from planetary remote sensing imagery, significantly surpassing the traditional limits of photogrammetric and photoclinometric methods. By integrating advanced deep learning approaches of super-resolution restoration (SRR) and monocular depth estimation (MDE), we transform single-view planetary orbital imagery into ultra-high-resolution digital terrain models (DTMs). Our Super-3D products, successfully validated using rover-derived 3D reference data from Mars, dramatically enhance the accuracy of topographic measurements, enabling detailed geological analyses of planetary surface features. Leveraging vision transformers and diffusion-based networks specially tuned for Mars and the Moon, we provide unmatched 3D precision for planetary scientists and mission planners.

Automated Multi-Resolution Image Co-registration

SAIIL offers advanced automated multi-resolution image co-registration services essential for accurately aligning planetary imagery from key imaging sensors of Mars and the Moon. Our pipeline utilises robust global least-squares fitting and mutual shape-adapted feature matching algorithms, achieving exceptional co-registration accuracy at the sub-pixel level of the coarser imagery. Additionally, we deploy cutting-edge deep-learning-based feature matchers pre-trained for Mars images, significantly enhancing matching speed and robustness, even for challenging terrains or poor image quality. This capability is critical for seamless integration of multi-instrument, spatial-temporal datasets, enabling comprehensive planetary studies and exploration planning.

Stereo 3D Reconstruction of Rover Imagery

SAIIL excels in detailed photogrammetric 3D reconstruction from Mars rover imagery, converting multi-view data collected by past and current Mars missions (Spirit, Opportunity, Curiosity, Perseverance) into coherent, high-fidelity 3D models. Using our state-of-the-art stereo reconstruction and structure-from-motion pipelines with bundle adjustment and precision 3D point cloud alignment, we produce wide baseline multi-resolution 3D products ready for visualisation and measurement. This work empowers planetary geologists to virtually explore Martian surface features, such as sedimentary layers and bedrock formations, performing precise measurements critical to scientific research and mission operations at unprecedented accuracy and detail.

Rapid AI Based 3D Estimation of Rover Imagery

SAIIL has developed deep-learning-based techniques for real-time planetary 3D reconstruction, delivering immediate and accurate 3D surface models directly from Mars rover imagery. Utilising advanced diffusion network architectures finely tuned for Mars rover datasets, our solution rapidly produces accurate and detailed 3D terrain models suitable for onboard rover navigation, operational planning, and immediate geological analysis. This capability significantly enhances real-time decision-making and operational efficiency, enabling Mars missions to adapt dynamically to the challenging planetary environment.

Dynamic Feature Detection and Tracking

SAIIL offers advanced capabilities for dynamic planetary surface feature detection and tracking using a combination of computer vision and deep learning algorithms, including image matching, change detection, region growing, and feature classification. Our solutions precisely identify and track transient geological phenomena such as Recurring Slope Lineae (RSL), dune migration, dust devil tracks, and seasonal frost changes on Mars. We offer tailored solutions and expert consultancy to optimise and retrain our dynamic feature detection and tracking system across diverse planetary environments.

Automated Crater and Rock Detection

At SAIIL, we provide state-of-the-art automated crater and rock detection services using advanced deep-learning architectures such as U-Net and vision transformer-based networks. Our robust AI-driven methods deliver rapid, precise, and scalable identification of planetary surface features, essential for geological mapping, landing-site assessments, and autonomous navigation planning. By employing customised neural networks trained on extensive planetary datasets, we ensure reliable detection performance across varied terrains and imaging conditions. Our automated workflows significantly streamline planetary surface analysis tasks, enabling scientists and mission teams to focus on interpreting geological contexts rather than manual feature extraction. We offer comprehensive solutions and technical support for mission-specific applications, ensuring optimal integration into planetary exploration projects.

The LISTER Toolbox

We have developed the LISTER (Lunar Monocular Image to Surface Topography Estimation & Reconstruction) toolbox, an advanced open-source processing system designed for high-resolution 3D lunar surface mapping using monocular orbital imagery. LISTER harnesses cutting-edge monocular depth estimation (MDE) techniques, integrating state-of-the-art deep-learning architectures, based on U-Nets, vision transformers, and diffusion networks, into a unified, automated workflow. This robust pipeline efficiently manages image tiling, MDE inference, adaptive elevation fitting, and seamless mosaicing, significantly simplifying the creation of accurate, large-scale lunar Digital Terrain Models (DTMs). Enhanced by comprehensive validation tools, automated quality assessments, and shadow-aware image enhancement routines, LISTER delivers highly accurate lunar surface models with unprecedented operational efficiency. It supports crucial lunar exploration tasks such as landing-site selection, hazard analysis, and scientific terrain mapping, offering planetary researchers and mission planners a powerful, reliable, and scalable solution for lunar topography estimation.

Large-area High-resolution 3D Mapping

We provide comprehensive solutions for large-area, high-resolution 3D mapping of planetary surfaces, essential for scientific exploration, robotic operations, and future human missions. Our integrated photogrammetric and deep-learning pipelines efficiently process multi-resolution datasets from key imaging instruments, generating detailed and consistent 3D basemaps across extensive regions of Mars. Our high-quality 3D products are actively supporting ESA’s ExoMars and NASA’s Mars2020 missions, demonstrating their utility for both scientific investigation and operational planning, and positioning SAIIL as a key contributor to next-generation planetary exploration missions.

Report abuse