How many hours does it take to locate a single red helmet in an 183-hectare wide area of rough mountain terrain? For the Italian National Alpine and Speleological Rescue Corps (CNSAS), the response is no longer in weeks or months, but hours through the marriage of artificial intelligence with drones. In July 2025, the intersection of human judgment and machine accuracy resulted in the identification of Nicola Ivaldo, a missing hiker of more than ten months on the north face of Monviso, the highest peak of the Cottian Alps. The rapidity and precision of the operation reveal an immense change in search and rescue tactics, with AI-based analysis now revolutionizing the previously painstaking, manually intensive process.
The rescue operation started with a daunting task: 2,600 frames captured by drones, one photo taken 50 meters above the ground for each, sweeping across a broad, hostile terrain. Normally, a data set like this would require painstakingly slow manual inspection, extending the timeframe for possible rescue or retrieval. But CNSAS made use of AI software that can identify colors and shapes, technology they had been developing for more than a year and a half. As explained by drone pilot Saverio Isola, It was the AI software that identified some pixels of a different color in the images taken on Tuesday, highlighting the helmet contrasting against the alpine landscape. The outcome was a drastic reduction in analysis time from a human timescale of months to a machine timescale of an afternoon.
This efficiency jump is based on innovations in convolutional neural networks (CNNs) and object detection architectures. Current systems, including faster region-based CNN (R-CNN) and single-shot detectors like YOLO, have empowered drones to process aerial imagery in real time even when objects are small, sparsely dispersed, or occluded by intricate backgrounds. As explained in more recent studies, multimodal data fusion integrating RGB images and depth maps or thermal images has further enhanced detection quality and reliability, especially under the adverse conditions of mountain search and rescue.
CNSAS methodology is exemplary for strict methodology. Each drone flight is carried out according to procedures agreed with ENAC, Italy’s civil aviation agency, thereby guaranteeing both safety and compliance with regulation when operating in alpine terrain. The workflow of the team is an exercise in efficiency: pictures are gathered over a five-hour period, and then quickly analyzed using AI to produce a shortlist of “suspicious spots.” Human judgment is still crucial four experienced mountain rescuers and two handpicked drone pilots analyze the output from the AI, check for confirmation on the ground, and coordinate with emergency services for the recovery efforts. It’s a human achievement, but without technology, it would have been an impossible mission. It’s a team success, Isola asserted.
Thermal imaging cameras are just one example of how AI is being married with high-end sensor payloads. CNSAS and other top search and rescue agencies are starting to deploy thermal imaging cameras in addition to standard RGB sensors. Thermal imaging, as investigated in recent research, is priceless for locating living victims in real-time, particularly at night or in low-visibility conditions. Thermal datasets can be retrained in algorithms like YOLOv5 to identify human heat signatures in relation to background noise, and tracking with Kalman filters sustains precise identification even in partially occluded or moving targets. Just like with still images, AI is also able to interpret thermal data and provide valuable information in just a few hours, Isola said, pointing to the increasing use of sensor fusion in contemporary rescue missions.
Still, there are technical hurdles. Drones have stringent payload restrictions, and atmospheric conditions like changing lighting, shadows, and poor weather can compromise image quality and detection accuracy. Researchers have countered by creating strong, lightweight neural network designs that can operate in real time on embedded devices such as the NVIDIA Jetson Xavier, so that advanced analytics can be conducted directly onboard the UAV. Such systems are tested against rigorous benchmarks, such as datasets containing high-resolution, annotated aerial imagery of living and non-living targets.
The impact on search and rescue is profound. By automating the preliminary review of massive datasets, AI enables human teams to concentrate their efforts where they are most valuable, lowering risk and enhancing the chances of saving lives. With CNSAS further developing its AI-drone approach, hope is obvious: that technology will not only speed up recoveries but also save lives by finding missing people before they perish. Nicola Ivaldo’s recovery is a testament to technological advances and also a glimpse of things to come: a future where machine intelligence and human expertise work in seamless partnership on the globe’s most daunting landscapes.
