Why time is the decisive factor in a search
In search and rescue operations, the count is in hours. The longer a person stays unfound, the larger the area that has to be swept and the higher the risks — hypothermia, dehydration, injury, disorientation. So the key metric for any search technology is the speed of area coverage while keeping detection reliable.
A classic ground search by a team is effective but slow: a line of people physically covers the terrain step by step, and difficult ground — deadfall, marsh, ravines — slows movement further still. At night, ground search either stops or sharply loses reliability.
An ordinary camera drone speeds up the flyover but moves the bottleneck to the person at the controls: the operator must continuously stare at the video stream and not miss a small figure among the foliage. Human attention is not limitless, and over large areas and long shifts, fatigue inevitably leads to misses.
This is where the thermal camera + onboard AI combination changes the picture. The drone covers the area quickly, the thermal camera makes a person visible even in foliage and darkness, and the AI takes on the most tiring part — the continuous analysis of every frame. It is the same principle of edge AI onboard a drone used in infrastructure inspection, only trained to recognize a person.
Thermal imaging + AI human detection
A thermal camera captures not light but the temperature of surfaces in frame. A human body is usually warmer than the surrounding forest, grass, or water, so on a thermal image it shows up as a contrasting spot even where, in ordinary light, the figure is fully hidden by shadow or foliage. This gives a search two important advantages: operation in the dark and the ability to isolate a person against a complex background.
But a thermal camera alone is not a solution. There are many warm objects in frame: animals, sun-heated rocks and trunks, cooling campfires, equipment. If all of this were flagged as a target, the operator would drown in false alarms. This is where AI human detection is needed.
The onboard computer-vision model is trained on the characteristic human signature — its shape, size, thermal profile, and movement pattern — and separates it from animals and inanimate heat sources. Instead of "here are a hundred warm dots," the operator gets "here are a few promising targets with confirming frames." Reducing the flow of false alarms here is not cosmetic but the condition under which an area search remains manageable at all.
The processing scheme fully matches other VOLKODAV scenarios: camera → Raspberry Pi-class onboard computer → computer-vision model → decision. Only the sensor changes (emphasis on the thermal channel) and the training set (a human signature instead of tower defects). We describe the same approach to finding thermal anomalies in the article on AI drone inspection of power lines — there the model looks for hot spots, here for people.
Why onboard AI is faster than an operator
The main advantage of onboard AI in a search is that it removes the limit of human attention. Let us compare two modes of a drone working over the same area.
| Parameter | Camera drone (manual review) | Thermal drone with onboard AI |
|---|---|---|
| Frame analysis | Operator's eyes, sequentially | Model, every frame at once |
| Visibility in the dark | Low | High (thermal channel) |
| Fatigue | Grows over time, leads to misses | Does not degrade |
| What the human gets | The whole stream to review | A short list of flags |
| Speed over area | Limited by review speed | Limited by flight speed |
| Works without a link | Depends on video relay scheme | Yes (edge, offline) |
The difference is fundamental. In manual mode, search speed is limited not by the drone but by how fast and attentively a human can review video. Onboard AI moves the analysis onto the aircraft itself: the model checks every frame in real time and brings in the operator only for suspicious areas. The area search runs in parallel with the flight, and the human concentrates on verifying a short list of targets rather than on reviewing the whole stream.
The low latency of edge processing also means that on detection the drone can immediately adjust its path — linger over the target, approach from a different angle, give the operator a better frame for confirmation — without waiting for a command over the network.
Working with no coverage and no GPS
The paradox of a search is that people most often get lost exactly where there is no link: deep in the forest, in the mountains, in marshes, far from settlements. Which means a search technology that depends on the cloud and the mobile network fails in precisely the conditions it is needed for.
The VOLKODAV core is built on edge processing: human detection runs on a Raspberry Pi-class onboard computer right in flight, without sending video to a server. The practical consequences for a search:
- Offline search. The drone recognizes a person with no cloud and no mobile network. Operation is not tied to a coverage zone — a decisive factor for remote terrain.
- Privacy. What goes out is not raw video but a flag and a confirming frame; sensitive footage is not pushed over open channels.
- Navigation autonomy. Onboard guidance in the VOLKODAV system orients by camera video and is not rigidly tied to GPS, so it stays operational where the satellite signal is weak or unstable.
The principle of autonomous operation without cloud or satellite is examined in more detail in the piece on edge AI on Raspberry Pi for drones, and the technical parameters of the core are gathered in the Specs section.
Scenarios: forest, field, night, water
Because the core is universal, search unfolds as a set of conditions, in each of which the thermal camera and the AI contribute in their own way.
Forest and rough terrain
The most common and the hardest scenario. Canopy and undergrowth hide a person from a regular camera, but the thermal contrast of the body helps isolate them in gaps and under the canopy. The AI keeps attention on the human signature and filters out warm animals.
Open field and steppe
Here the area is vast and landmarks are few. Coverage speed comes to the fore: the drone flies over large expanses in minutes, while the AI flags any thermal targets resembling a person so the team does not sweep the field wholesale.
Night search
At night a thermal camera works even better than by day: the background cools and the contrast of a warm human body against the surroundings increases. Where a ground night search almost grinds to a halt, a drone with a thermal channel keeps searching effectively.
Shoreline and water
At the boundary of water and land, a person stands out sharply by thermal signature against the cold water surface. This eases detection along shores, on shoals, and in marshy areas.
An important caveat: no technology guarantees detection in all conditions. Dense fog, heavy rain, and very thick canopy reduce the effectiveness of the thermal channel, as with any such sensor. So the drone is a force multiplier for the search team, not a replacement for it, and we honestly do not promise a fixed result.
The drone as part of the search team
An AI drone fits into the work of a search team as a means of fast coverage and prioritization, not as a standalone "finder." A practical scheme:
- A fast flyover of the probable zone right after the task is set — the drone covers an area unreachable by a foot team in the same time.
- Automatic flagging — every flyover yields a list of thermal targets classified as promising, with coordinates and frames.
- Route prioritization — ground groups head first to the flagged points rather than sweeping everything indiscriminately.
- Verification and assistance by people — confirmation, evacuation, and aid always stay with rescuers; the AI only narrows the search zone.
Because it uses the same universal core as other VOLKODAV tasks, no separate development for search is required — the training data and the payload change. How one and the same AI module covers both civilian and other tasks is discussed in the article "One AI Module — Two Tasks." To discuss use under your conditions, see the Contact section.
Conclusion
Search and rescue with AI and thermal drones speeds up an operation where time decides everything. The thermal camera makes a person visible in foliage and darkness, while the onboard AI analyzes every frame in real time — faster than an operator can review video — and flags promising targets, filtering out false heat sources. Edge processing allows searching offline, with no network coverage and with navigation autonomy, that is, exactly where people get lost most often. The drone reinforces the search team, while the final decision stays with people. For more on applications, see the Use cases section; for the operating principle, see How it works.