8 Overwatch Imaging is known for its intelligent multi-sensor payloads for airborne survey operations, but it recently developed an AI-driven software called Automated Sensor Operator (ASO) for automating image collection and analysis tasks for common video gimbal platforms. A key part of what sets ASO apart from other automated image-collection and analysis systems is that it has been designed to work at the edge, onboard the uncrewed platform. This ensures maximum fidelity and quality of data, compared with systems that work on the ground or in the cloud, where data often has to be compressed or it can be lost. “Taking advantage of all that data at the same time in one processing step, before it gets compressed or transmitted, simply allows us to find more in the data stream compared to software that works later and downstream of where ASO works,” said Greg Davis, founder and CEO of Overwatch Imaging. Overwatch Imaging provides ASO on a small, Nvidia GPU-based device, which the systems integrator connects first to their imaging system and then to either the comms system for the ground operations crew or to the autopilot. “By operating in real time at the edge, we’re able to automate an operational ‘closed-loop’ around intelligent detections of mission-critical items,” Davis said. Typical automatic target recognition or machine learning-based object-detection models for UAVs do work well, but Davis comments that these rarely do more than place an annotated bounding box around the recognised object and flag an alert, leaving the loop “open”. As ASO controls payload sensors in an automated fashion upon detecting and recognising an object, it can be configured to have the gimbal or aircraft react intelligently to the finding. This could mean maintaining focus on the target or returning periodically to track its movements, or looking around the area to see what approaches. Either way, this reduces the burden on the operating crew to physically intervene to trigger a reaction from the UAV or sensor payload. Developing ASO was partially made possible through what Davis calls Overwatch Imaging’s “lake” of data: hundreds of terabytes of multi-spectral, high-resolution, air-to-ground imagery, showing countless relevant scenes of interest to UAV operators. “That enables us to train models, and not just first-order model training, because we’ve captured a huge number of corner or edge case examples, so we can train our models against those, which has helped us make the system more useful,” he said. “Being able to discriminate in many cases between ‘subtle targets’ and ‘non-targets very similar in appearance to targets’ helps the case for operators to be assigned very different tasks, most likely reducing their workload to tasks appropriate for human intelligence and skillsets, within an uncrewed operation.” The importance of this stems from one of the biggest deterrents for survey organisations thinking of adopting UAVs, which is that doing so often entails enlarging field-survey crews, or making them work harder or undertake considerable training. Automating image analytics and responses about the sensor stands to make surveyors’ lives easier by using UAVs. By building its own sensors, Overwatch Imaging has populated its data lake with images across multiple spectral bands, and can rapidly update its models in isolation from one another, and from its core architectural software. That also means the ASO is fully functional with imaging cameras operating in the UV, visible, NIR, SWIR, MWIR and LWIR bands, along with fullmotion video. “We anticipate search-and-rescue operations being performed by the different military branches in dangerous and obscure conditions, and coast guard search and rescue, where personnel are racing against time to perform the hugely difficult task of finding souls lost at sea – as one especially vital initial application of ASO,” Davis added. AI Automating image collection and analysis with AI Platform one December/January 2025 | Uncrewed Systems Technology ASO is designed to work at the edge, onboard the uncrewed platform, ensuring maximum quality of data
RkJQdWJsaXNoZXIy MjI2Mzk4