Comprehensive Vision-Based AI Capabilities: SAIIL’s low-power AI Processing Units (AIPUs) robustly support a comprehensive range of vision-based tasks using state-of-the-art deep learning networks and advanced computer vision algorithms. Our capabilities include 2D/3D matching and mosaicing, object and change detection, segmentation and masking, object and scene tracking, classification, super-resolution, 3D reconstruction, video compression, and AI analytics. This extensive suite of vision services ensures customers can seamlessly deploy sophisticated visual intelligence onboard their platforms, addressing diverse application requirements with exceptional accuracy and real-time responsiveness.
Broad Sensor Integration: Our AIPUs are designed to integrate with a wide range of imaging and sensing technologies, including thermal infrared (TIR) cameras, stereo vision cameras, wide-angle imaging systems, LiDAR sensors, and other specialised devices. This broad compatibility enables customers to implement advanced, multi-sensor fusion workflows onboard their satellites, UAVs, robots, and handheld devices, significantly enhancing situational awareness, autonomy, and mission effectiveness across numerous domains.
Accelerated AI Model Performance: To maximise efficiency and performance, our AI processing units leverage cutting-edge AI acceleration techniques, including (1) neural architecture search to systematically design lightweight optimisations for existing networks; (2) post-training quantisation techniques to significantly reduce GPU memory footprint and enhance computational speed; and (3) explainable knowledge distillation to efficiently transfer knowledge from complex models to simplified networks suitable for onboard inference. These approaches ensure our AI models deliver greater than 99% of baseline accuracy while operating 2–3 times faster and utilising over 50% less memory, critical for achieving real-time analytics and processing in power- and resource-constrained environments.
Space-Tested Reliability: SAIIL’s AIPUs are built on industrial-grade computing hardware with demonstrated reliability in rigorous radiation environments. Extensive proton-beam testing campaigns confirm tolerance of significant radiation doses, and robust onboard watchdog routines actively detect and mitigate single-event effects. Our dual-stage supervisory system continuously monitors memory integrity and autonomously initiates rapid resets when necessary, ensuring uninterrupted, dependable performance even in demanding space conditions.
Seamless Platform Integration and Support: We offer comprehensive technical support to facilitate the seamless integration of our low-power AIPUs into various platforms, including small satellites, UAVs, robotic systems, handheld imaging devices, VR headsets, and scientific imaging systems. Our expert team actively collaborates with partners on the AIPU integration process, ensuring rapid deployment and immediate operational benefits. SAIIL is continually seeking new partnerships to expand the adoption of our AIPU technology, empowering enhanced real-time vision processing and AI-driven analytics across a wide range of applications.
We offer advanced real-time cloud motion vector (CMV) / wind speed estimation systems based on integrated thermal infrared (TIR) imaging and AIPU. Our solution uses robust deep-learning optical flow networks to provide sub-pixel-level horizontal wind measurements at cloud-base in real-time. Tested extensively by our project partners at the Chilbolton Facility for Atmospheric Radar Research (CFARR), our AIPU demonstrated reliable real-time performance and accurate retrievals validated against lidar measurements. This innovative technology is crucial for meteorological forecasting, aviation safety through wind-shear detection, and climate research. We provide customisable AIPU solutions, technical integration, and ongoing support for clients in meteorology, aviation, renewable energy, and climate research communities.
Cloud Motion Tracking
Wind Speed Estimation (u)
Wind Speed Estimation (v)
Our onboard AI video compression and FPV video enhancement solutions significantly enhance operational capabilities of Beyond Visual Line-of-Sight (BVLOS) UAV missions. Leveraging state-of-the-art deep-learning methods, our AIPUs execute efficient video compression alongside advanced real-time analytics onboard UAV platforms. This dramatically reduces transmission bandwidth requirements (by up to 80%) without compromising critical visual details. Our advanced AI enhancement techniques further reconstruct transmitted video to ultra-high-definition clarity, crucial for precise surveillance, industrial inspection, and search-and-rescue missions. SAIIL provides customised solutions, extensive UAV platform integration support, and seeks partnerships to optimise real-time video analytics capabilities for UAV operators.
Onboard AI Video Compression
AI enhanced UAV video (after transmission)
Scene tracking using AI enhanced video
SAIIL’s handheld imaging technology for real-time 3D reconstruction offers a transformative solution for infrastructure inspection tasks. Our fully optimised monocular depth estimation (MDE) networks process ultra-high-definition images in realtime onboard the low-power AIPU, enabling immediate detection and assessment of structural anomalies. Demonstrated effectively in railway inspections, our solution can rapidly identify defects, measure precise dimensions, and produce detailed 3D models directly onsite. Beyond railways, our versatile handheld 3D imaging device is equally suited to inspection and monitoring applications in tunnels, bridges, building facades, and heritage conservation, delivering unprecedented accuracy, efficiency, and safety benefits for infrastructure managers.