# How does drone detection work?

> Radar, radio, microphones and cameras each find drones differently. What each one misses, and how a camera system keeps a drone in view.

Page: https://www.whitehatdesignstudio.com/guides/how-drone-detection-works/
Published by: WhiteHat Design Studio
Last updated: 7 November 2026

The short answer

Drone detection finds a drone with one or more sensors: radar, a radio receiver, microphones or a camera. Each one has a blind spot, so the strongest systems combine them. A camera system such as SKYLOCK runs a detection model on every frame, predicts where the drone is going and turns a gimbal to keep it in view.

By the engineers at WhiteHat Design Studio. Published 7 November 2026.

![The SKYLOCK camera gimbal from behind, out of focus in the foreground of a garden, with a small drone in a clear sky above the trees.](https://www.whitehatdesignstudio.com/_astro/skylock-gimbal.BC1TUiGl_ZJjn62.jpg)

The SKYLOCK gimbal outdoors, with a drone above the trees.

## The four ways to find a drone

Four kinds of sensor do nearly all the work. A review published in the journal Sensors in 2024 compared radar, sound, cameras and radio side by side. The table follows its findings.

| Sensor | Good at | Falls short when |
| --- | --- | --- |
| Radar | Long range, in any weather | The drone is small, so its echo is small. Active radar also needs a transmission licence. |
| Radio receiver | Picking up the link between a drone and its pilot | The drone transmits nothing at all |
| Microphone | Low cost | The drone is far away. The review puts acoustic range under 500 m. |
| Camera | Showing what the object is | There is no clear line of sight, or fog, dust or cloud is in the way |

The review’s conclusion is that systems combining several sensors give the stronger results. Each sensor covers a gap in another: radar can tell a camera where to look, and the camera can show what the radar found.

## What a camera adds

Radar and radio can tell you that something is there. A camera shows you what it is, and gives a person something to judge. That is why SKYLOCK, the drone detection and tracking system we built, centres on a camera on a gimbal and a console an operator can read.

The hard part is size. A small drone at a distance is a few pixels against a bright sky, and it has to be found in every frame, quickly enough to keep up with it.

## How a camera keeps a drone in view

A camera tracking system is a loop. SKYLOCK runs it in four stages.

1. **Get the video.** The camera on the gimbal streams to a ground control station over RTSP, a protocol for setting up and controlling the delivery of real-time video. Every stage of that path adds delay, so the station is always looking slightly into the past.
2. **Find the drone.** A YOLO26 detection model from Ultralytics marks each drone it finds in a frame, with a score for how sure it is. We run the model through NVIDIA TensorRT, which compiles it for low latency on the hardware it runs on.
3. **Predict its path.** A detector misses frames, and its boxes jump about. A Kalman filter keeps a running estimate of where the drone is and how it is moving. It predicts the next position, then corrects that prediction with each new detection, giving more weight to whichever is more certain. SKYLOCK’s console marks the predicted point on screen, just ahead of the drone.
4. **Keep it in view.** The gimbal turns to bring the drone back to the centre of the view. SKYLOCK’s console offers a joystick, pointer tracking and auto track. In auto track the gimbal follows the drone on its own, with no operator needed.

## Where it gets hard outside the lab

- **Small and far away.** The further the drone, the fewer pixels it fills, until it is hard to tell from noise. Ultralytics designed YOLO26’s training to keep labels on small objects, which helps, but no model can find what the camera did not capture.
- **Things that move like drones.** Birds and other small moving objects cross the same sky. The detector’s score, the predicted path and a person at the console all help tell them apart.
- **The sky changes.** Cloud, glare and a low sun change what a frame looks like from one minute to the next, so test in the light you will really have.
- **Delay.** Video, detection and the gimbal’s own movement each take time. Part of the prediction step’s job is to cover that gap.
- **Losing it.** When a drone passes behind trees, the system has to keep going on its prediction and find the drone again when it reappears.

## What to ask any supplier

A range or an accuracy figure means little on its own. Ask for each number with its conditions:

- Which drone, at what size, and at what distance?
- In what light and weather?
- How often does it raise a false alarm, per hour, and what set it off?
- How long from a drone appearing to the first detection?
- On which computer, and at how many frames a second?
- What does it do when it loses the drone?

We hold ourselves to the same test. We have not published a detection range, an accuracy figure or a response time for SKYLOCK, because a number without its conditions tells you nothing. They will appear on the [SKYLOCK case study](https://www.whitehatdesignstudio.com/work/skylock/) once they are measured, with the conditions they were measured in.

## Where the answer comes from

- [SKYLOCK](https://www.whitehatdesignstudio.com/work/skylock/) Finds drones in live camera feeds, predicts where each one is going and keeps a gimbal camera on it, from the operator's console or on its own.

Practices

- [Hardware](https://www.whitehatdesignstudio.com/services/hardware/) Machines that survive the real world.
- [Software](https://www.whitehatdesignstudio.com/services/software/) Software that runs the operation.
- [AI](https://www.whitehatdesignstudio.com/services/ai/) Intelligence that runs where the work is.

A hardware, B software, C AI

Drawn from the project descriptions and what the console shows on screen.

## On film

Finding a drone in the live camera feed and keeping the gimbal on it. A clip from [the log](https://www.whitehatdesignstudio.com/log/), without sound.

[SKYLOCK, following a drone in auto track. Video without sound, 21 seconds.](https://www.whitehatdesignstudio.com/media/clips/skylock-tracking.mp4)

0:21, no soundThe SKYLOCK console in auto track. A small drone crosses a pale sky; a red box marks it with its detection score, and a small circle shows where the software predicts it is going. An inset then shows the camera gimbal on its tripod outdoors as it follows. By the end the drone sits at the centre of the view.

## Questions people ask next

### Can a camera detect a drone at night?

A normal camera needs light, so a dark sky hides a drone unless the drone is lit. Thermal cameras see heat rather than light, which is why they are used after dark. Our published SKYLOCK recording is in daylight.

### Is one kind of sensor enough?

It can be, if you know its blind spot and can live with it. A camera will not see through fog, and a radio receiver will not find a drone that transmits nothing. Where it matters, a second kind of sensor covers the first one's gap.

### What is the difference between detection and tracking?

Detection finds the drone in one frame. Tracking follows the same drone from frame to frame, so the system knows where it is going and can keep a camera on it.

## Sources

- [Famili and others, Securing your airspace, detection of drones trespassing protected areas, Sensors, 2024](https://pmc.ncbi.nlm.nih.gov/articles/PMC11013887/)
- [Abu Zitar and others, A review of UAV visual detection and tracking methods, 2023](https://arxiv.org/abs/2306.05089)
- [Ultralytics, YOLO26 documentation](https://docs.ultralytics.com/models/yolo26/)
- [NVIDIA, TensorRT](https://developer.nvidia.com/tensorrt)
- [IETF, RFC 7826, the Real-Time Streaming Protocol, version 2.0](https://www.rfc-editor.org/rfc/rfc7826.html)
- [Wikipedia, Kalman filter](https://en.wikipedia.org/wiki/Kalman_filter)

## More guides

- [STM32 vs ESP32: which should your product use?](https://www.whitehatdesignstudio.com/guides/stm32-vs-esp32/)
- [Edge AI or cloud AI: where should your model run?](https://www.whitehatdesignstudio.com/guides/edge-ai-vs-cloud-ai/)
- [Visual SLAM vs LiDAR SLAM for indoor robots](https://www.whitehatdesignstudio.com/guides/visual-slam-vs-lidar-slam/)

[All the guides](https://www.whitehatdesignstudio.com/guides/)

## Tell us what has to work.

One sentence is enough. An engineer will reply.

[Talk to an engineer](https://www.whitehatdesignstudio.com/contact/) [WhatsApp an engineer](https://wa.me/923112721007?text=Hello%20WhiteHat%20Design%20Studio%2C%20I%20would%20like%20to%20talk%20to%20an%20engineer%20about%20my%20project.)