Visual-Inertial Odometry for Drones

Visual-Inertial Odometry, commonly known as VIO, is an increasingly important navigation technology for drones. It allows an aircraft to estimate its movement and position by combining information from cameras with measurements from an Inertial Measurement Unit (IMU). This makes VIO particularly valuable when GNSS signals are unavailable, unreliable, degraded or temporarily obstructed.

Traditional drone navigation depends heavily on satellite positioning systems such as GPS, Galileo, GLONASS or BeiDou. These systems work extremely well in open outdoor environments, but their performance can deteriorate inside buildings, beneath bridges, in tunnels, underground mines, around tall structures, beneath dense vegetation and in other environments where satellites cannot be observed reliably.

VIO provides another source of navigation information. Cameras observe how visual features move between consecutive images while the IMU measures acceleration and rotation. By combining these measurements, the drone continuously estimates how it has moved through three-dimensional space.

This capability supports applications including indoor inspection, infrastructure inspection, warehouse operations, underground mining, tunnel mapping, autonomous robotics, construction, emergency response, GPS-denied navigation and autonomous drone operations.

However, VIO should not be confused with an absolute positioning system. It estimates movement relative to an initial position and can accumulate error over time. Lighting, visual texture, camera configuration, IMU quality, vibration, calibration and environmental conditions can all influence performance.

The strongest VIO systems therefore combine high-quality cameras, reliable inertial sensors, precise calibration, robust feature tracking, sensor fusion, environmental awareness and additional positioning technologies where available.

What Is Visual-Inertial Odometry?

Visual-Inertial Odometry estimates the movement of a drone by combining visual information from one or more cameras with measurements from an IMU.

The visual system observes identifiable features within the environment. These might include corners, edges, patterns, surface textures or other distinctive details. As the drone moves, the apparent position of these features changes between camera frames.

Software analyses these changes to estimate movement.

At the same time, the IMU measures linear acceleration and angular velocity. These measurements provide rapid information about how the aircraft is moving and rotating.

Combining the two data sources produces a more robust estimate than either system would normally provide independently.

Why Drones Need VIO

GNSS has transformed drone operations because it provides convenient global positioning. However, there are many environments where GNSS alone is insufficient.

A drone flying into a warehouse may lose satellite reception. A drone inspecting beneath a bridge may experience degraded positioning. Underground mines and tunnels may have no satellite reception at all.

Urban environments can also create multipath effects as satellite signals reflect from buildings.

VIO allows the drone to continue estimating its motion when satellite positioning becomes unreliable.

This can support stable flight, navigation, mapping and autonomous operation.

For this reason, VIO is becoming increasingly important as drones move beyond open-air operations and into complex environments.

How VIO Works

A VIO system continuously receives information from its cameras and IMU.

The camera provides observations of the surrounding environment.

The IMU provides high-frequency measurements of aircraft movement.

Computer-vision algorithms identify and track visual features across successive images. The system estimates how the camera must have moved to produce the observed changes.

Meanwhile, the inertial system predicts movement based on acceleration and rotation.

Software combines these estimates mathematically.

The camera helps correct the accumulating errors of inertial navigation, while the IMU provides rapid motion information between camera frames.

The result is a continuously updated estimate of the drone’s position, velocity and orientation.

Visual Odometry

Visual odometry is the process of estimating motion using sequential camera images.

Suppose a camera observes a corner of a building.

As the drone moves forward, that corner changes position within the image.

By tracking many features simultaneously, software can estimate how the camera moved between frames.

When this process is repeated continuously, the drone builds an estimated trajectory.

However, visual odometry alone can struggle during rapid motion, poor lighting or when the camera sees few distinctive features.

The IMU helps compensate for these limitations.

The Role of the IMU

The IMU normally contains accelerometers and gyroscopes.

Accelerometers measure specific force, which contributes to estimating changes in velocity and position.

Gyroscopes measure angular velocity.

These measurements are produced at much higher frequencies than typical camera frames.

This means the IMU can detect rapid movements between images.

However, inertial sensors contain small measurement errors and biases.

When these errors are integrated over time, they can create substantial drift.

Visual measurements therefore provide important corrections.

Why Camera and IMU Integration Works

Cameras and IMUs have complementary strengths.

Cameras provide information about movement relative to visible features, but their measurements arrive comparatively slowly and depend on environmental conditions.

IMUs operate extremely quickly and do not require visible environmental features.

However, inertial errors accumulate rapidly.

Combining both sensors allows each technology to compensate for weaknesses in the other.

This is the fundamental principle behind VIO.

The quality of the fusion algorithm is therefore just as important as the quality of the individual sensors.

Monocular VIO

A monocular system uses a single camera.

This has advantages in size, weight, cost and processing requirements.

However, a single camera cannot directly determine absolute scale from images alone without additional information.

The IMU helps resolve scale because it measures physical acceleration.

Monocular VIO can therefore provide useful three-dimensional motion estimation.

However, performance depends heavily on good calibration and robust visual tracking.

Stereo VIO

Stereo systems use two cameras separated by a known distance.

By comparing the images from both cameras, the system can estimate depth.

This provides additional geometric information.

Stereo VIO can be particularly useful for obstacle avoidance and indoor navigation.

However, two cameras increase payload size, weight and computational requirements.

The cameras also need accurate calibration relative to one another.

For many autonomous drones, the additional depth information can justify this complexity.

Multi-Camera VIO

More advanced drones may use several cameras facing different directions.

This increases the field of view available to the navigation system.

If one camera loses useful features, another may still see sufficient geometry.

Wide-angle or fisheye cameras are commonly used because they observe large areas of the environment.

Multi-camera systems can therefore improve robustness in complex spaces.

However, every camera must be accurately calibrated and synchronised.

The processing requirements also increase substantially.

Camera Field of View

A wide field of view can benefit VIO because the system sees more environmental features.

It also allows features to remain visible for longer as the drone moves.

This can improve tracking.

However, wide-angle lenses introduce optical distortion.

VIO software must model this distortion accurately.

Fisheye cameras can provide excellent navigation information when correctly calibrated.

The camera should therefore be evaluated as part of the complete navigation system rather than by image resolution alone.

Feature Detection

Visual odometry depends on identifying features that can be recognised across multiple frames.

Corners, textured surfaces and strong visual patterns are particularly useful.

A brick wall, for example, may contain many trackable features.

A completely smooth white wall may contain very few.

Algorithms identify candidate features and assign mathematical descriptions to them.

The software then attempts to find the same features in subsequent images.

The quality and distribution of these tracked features influence the navigation estimate.

Feature Tracking

Once features have been identified, the system follows them as they move through the image.

Their apparent motion contains information about the camera’s movement.

Tracking many features provides redundancy.

If several features disappear, others may remain.

However, moving objects can complicate the process.

A person walking through the scene should not be treated as a fixed navigation reference.

Advanced algorithms therefore attempt to distinguish stable environmental features from dynamic objects.

Optical Flow

Optical flow describes the apparent movement of pixels or image features between frames.

It can provide information about relative motion.

Some drones use downward-facing optical-flow cameras for low-altitude position holding.

VIO extends this concept by combining visual motion with inertial measurements and three-dimensional geometry.

Optical flow alone should therefore not be considered identical to full visual-inertial odometry.

VIO generally provides a more complete estimate of motion and orientation.

Visual-Inertial Sensor Fusion

Sensor fusion is the process of combining camera and IMU information into one navigation estimate.

Different systems use different mathematical approaches.

Filtering methods may continuously update an estimated state as new measurements arrive.

Optimisation-based systems may evaluate multiple observations together.

The exact algorithm varies between manufacturers and research systems.

For users, the important consideration is whether the complete system provides reliable localisation under the expected operating conditions.

Camera-IMU Synchronisation

Timing is critical.

The drone may rotate rapidly.

If a camera image is associated with IMU measurements from the wrong moment, the system can calculate an incorrect trajectory.

Even small timing offsets can become significant during fast movement.

Professional VIO systems therefore require precise camera-to-IMU synchronisation.

Hardware triggering and accurate timestamps are commonly used.

A high-quality camera and IMU can still produce poor navigation if their timing relationship is incorrect.

Camera-IMU Calibration

The software must know the exact physical relationship between the camera and IMU.

This includes their relative position and orientation.

These parameters are sometimes called extrinsic calibration.

Camera properties such as focal length and lens distortion must also be known.

These are intrinsic parameters.

Incorrect calibration can create systematic navigation errors.

Mechanical mounting is therefore important because movement between the camera and IMU can invalidate the calibration.

Scale Estimation

A camera observes geometry through images but does not necessarily know the physical scale of the scene.

Inertial information helps provide this scale.

The system knows that measured acceleration represents physical movement.

Combining this with visual motion allows the VIO algorithm to estimate distances in real-world units.

This is one of the reasons inertial integration is particularly valuable for monocular systems.

However, inaccurate IMU calibration can influence scale estimation.

Position Estimation

VIO produces an estimate of the drone’s changing position.

This is normally expressed relative to the position where the VIO system was initialised.

For example, the system may consider its starting point to be coordinates 0, 0, 0.

As the drone moves, it estimates its displacement from that location.

This is different from GNSS, which provides coordinates relative to a global reference system.

Additional information is therefore required if the VIO trajectory needs to be aligned with real-world geographic coordinates.

Orientation Estimation

VIO also estimates aircraft orientation.

This includes roll, pitch and yaw.

Gravity provides a strong reference for roll and pitch when combined with inertial measurements.

Heading can be more challenging.

Visual geometry can help estimate relative heading changes.

However, without an external reference, the system does not automatically know geographic north.

GNSS heading, magnetometers or other references may be integrated when absolute heading is required.

Velocity Estimation

The system continuously estimates velocity.

This is important for flight control.

A drone needs to understand whether it is moving even when the pilot commands it to hover.

VIO can detect this motion relative to the surrounding environment.

The flight controller can then make corrections.

This is one reason visual positioning systems allow many modern drones to maintain stable hover indoors.

Drift

VIO is an odometry technology.

This means small errors can accumulate as the drone travels.

This is known as drift.

A small positional error during each movement estimate may eventually become a significant difference between the estimated and actual position.

The amount of drift depends on sensors, algorithms, environment and trajectory.

Additional localisation sources can be used to correct it.

VIO should therefore not automatically be treated as a permanent replacement for absolute positioning.

VIO Versus GNSS

GNSS provides global coordinates.

VIO provides relative movement estimates based on local observations.

The technologies are complementary.

Outdoors, GNSS can provide global reference while VIO contributes smooth local motion information.

When the drone enters a building and loses GNSS, VIO can continue tracking movement.

When reliable GNSS returns, the global reference can help constrain accumulated drift.

Hybrid navigation therefore provides greater resilience than depending entirely on either system.

VIO Versus GPS

GPS is one satellite constellation within the broader GNSS category.

In common drone terminology, GPS and GNSS are often used interchangeably, but professional systems may use multiple constellations simultaneously.

VIO does not depend on satellite signals.

This gives it a major advantage indoors and underground.

However, VIO does not inherently provide latitude, longitude or globally referenced altitude.

The two technologies solve different parts of the navigation problem.

VIO Versus Optical Flow

Basic optical-flow systems generally estimate motion by tracking visual patterns, often using a downward-facing camera.

They are commonly used for low-altitude hover.

VIO adds inertial measurements and usually provides more complete three-dimensional state estimation.

This makes it suitable for more advanced navigation.

However, optical flow can still be an important component within a wider navigation architecture.

The terminology used by manufacturers should therefore be examined carefully.

VIO Versus Visual SLAM

Visual-Inertial Odometry and Visual SLAM are closely related but not identical.

VIO primarily estimates incremental motion.

Visual SLAM also builds a map and attempts to recognise previously visited areas.

This recognition can provide loop closure.

Loop closure allows accumulated drift to be corrected.

A system may therefore use VIO for immediate real-time motion estimation while a SLAM layer performs longer-term map optimisation.

The two technologies are frequently combined.

VIO Versus LiDAR SLAM

LiDAR SLAM uses laser measurements rather than relying primarily on camera imagery.

LiDAR has advantages in darkness because it actively measures the environment.

Visual systems can be lighter and less expensive.

Cameras also provide rich appearance information.

The strongest autonomous platforms may combine both.

LiDAR can provide geometric structure while VIO provides high-rate visual-inertial motion information.

Sensor diversity improves resilience.

GNSS-Denied Navigation

One of the most important applications of VIO is navigation without reliable GNSS.

This includes indoor, underground and structurally obstructed environments.

The drone can continue estimating its motion using local visual information.

However, GNSS-denied does not automatically mean navigation is guaranteed.

VIO has its own environmental requirements.

Darkness, smoke, dust or visually featureless surfaces can reduce performance.

A resilient drone should therefore understand the confidence of its navigation solution.

Indoor Drone Navigation

Indoor drones cannot normally rely on GNSS.

VIO provides a practical alternative.

Warehouses, factories, hangars and commercial buildings often contain sufficient visual features for reliable tracking.

The drone can use VIO for position holding and autonomous route following.

However, lighting can vary dramatically.

Bright windows, dark corners and reflective surfaces may create difficulties.

Navigation cameras should therefore be designed for a wide range of lighting conditions.

Warehouse Operations

Warehouses are increasingly using autonomous robots and drones.

VIO can help drones navigate aisles and storage areas.

Cameras observe racks, walls and other structural features.

The system can be combined with barcode, RFID or inventory cameras.

However, long repetitive aisles may create visual ambiguity.

SLAM, LiDAR or artificial landmarks can improve navigation reliability.

VIO provides movement estimation but does not inherently know which shelf contains a particular product.

Industrial Inspection

Industrial facilities contain complex structures where GNSS may be weak.

VIO can help drones navigate around pipes, machinery, tanks and internal structures.

This can support visual, thermal, gas or NDT inspection.

The navigation system provides positioning while specialist payloads collect inspection information.

However, visual navigation cameras should not automatically be assumed to provide inspection-quality imagery.

Navigation and inspection cameras may have very different requirements.

Tunnel Inspection

Tunnels are natural applications for VIO because GNSS is unavailable.

Cameras can track structural features along walls and ceilings.

The IMU provides rapid motion information.

However, tunnels can also be challenging because they may contain repetitive geometry and poor lighting.

Artificial illumination may therefore be required.

Combining VIO with LiDAR can provide additional robustness for long tunnel missions.

Underground Mining

Underground mines present similar navigation challenges.

GNSS is unavailable and the environment may contain dust, darkness and repetitive tunnel structures.

VIO can contribute to localisation, especially when integrated with LiDAR.

However, dust can reduce visual contrast.

The drone’s lighting system therefore becomes part of the navigation architecture.

Mine environments should be tested carefully rather than assuming a system that works indoors will perform identically underground.

Beneath Bridges

GNSS can become degraded beneath bridges and large structures.

VIO can maintain local motion estimation during the temporary loss.

This allows smoother inspection around bridge decks and structural components.

When the aircraft returns to open sky, GNSS can re-establish the global reference.

This transition between navigation sources is an important capability for infrastructure inspection drones.

Urban Canyons

Tall buildings can block satellite visibility and create multipath.

VIO can help maintain local navigation between buildings.

Visual features are usually abundant in urban environments.

However, vehicles and pedestrians create dynamic objects.

The navigation algorithm should prioritise stable environmental features.

Combining VIO with GNSS, LiDAR and other sensors can improve resilience in complex cities.

Navigation Beneath Vegetation

Dense tree canopy can degrade GNSS reception.

VIO may provide additional relative navigation information.

Tree trunks and nearby terrain can provide visual features.

However, vegetation moves in wind.

Leaves and branches can therefore create unstable references.

Lighting beneath canopy may also vary rapidly.

VIO can support forest navigation but should normally form part of a broader sensor-fusion system.

Construction Sites

Construction environments frequently contain areas where GNSS performance changes as buildings develop.

A drone may begin outside with satellite positioning and then fly inside a partially completed structure.

VIO can maintain localisation through this transition.

The system can support progress monitoring and inspection.

However, construction sites change continuously.

Features that existed during one mission may disappear before the next.

Persistent mapping therefore requires robust map management.

Confined Spaces

VIO can support drones operating inside tanks, vessels and other confined environments.

The aircraft uses nearby visual features to estimate movement.

However, confined spaces can contain smooth surfaces with little texture.

Lighting may also be poor.

Metallic or reflective surfaces can complicate visual tracking.

Combining cameras with LiDAR or other ranging sensors can provide greater robustness.

Emergency Response

VIO can support drones entering buildings during emergency operations where GNSS is unavailable.

The aircraft may provide visual or thermal information to responders.

However, smoke, dust and darkness can significantly degrade camera-based navigation.

VIO should therefore not be treated as universally reliable in emergency environments.

LiDAR, radar or other complementary sensors may be valuable.

The drone supports situational awareness while trained responders remain responsible for operational decisions.

Search and Rescue

Search-and-rescue drones may use VIO to enter caves, buildings or other areas without GNSS.

The navigation system allows the aircraft to estimate its path.

This can be combined with thermal cameras, RGB cameras or communications payloads.

However, detecting a thermal or visual candidate does not confirm identity or condition.

VIO enables the drone to navigate; it does not interpret the medical or rescue significance of observations.

Low-Light Operation

Camera-based navigation depends on sufficient visual information.

In low light, image noise increases and exposure times may become longer.

Motion blur can then reduce feature tracking.

Navigation cameras may use high-sensitivity sensors and wide-aperture lenses.

Integrated illumination can also help.

However, lighting must cover the camera’s field of view without creating excessive glare.

Low-light performance should be tested under realistic operating conditions.

Complete Darkness

Conventional visible-light VIO cannot operate effectively in complete darkness without illumination.

The drone can carry LEDs or infrared illumination if the cameras are sensitive to the corresponding wavelengths.

Alternatively, LiDAR can provide geometry independently of ambient visible light.

This is why multi-sensor navigation is particularly useful for underground and confined-space drones.

No single sensor works optimally in every environment.

Motion Blur

Rapid movement can blur camera images.

This makes visual features harder to track.

Fast shutter speeds reduce blur but require more light.

Global-shutter cameras can also be advantageous because the entire image is captured at essentially the same instant.

Rolling-shutter cameras capture different rows at slightly different times, which can distort images during rapid motion.

VIO algorithms can model some of these effects, but suitable camera hardware remains important.

Global-Shutter Cameras

Global-shutter sensors are often preferred for high-performance VIO.

They reduce geometric distortion during rapid aircraft movement.

This makes feature tracking more consistent.

However, global shutter alone does not guarantee strong navigation.

Resolution, sensitivity, frame rate, field of view, calibration and synchronisation all matter.

The camera should be selected as part of the complete VIO system.

Frame Rate

Higher camera frame rates reduce the amount of movement between consecutive images.

This can improve tracking during fast flight.

However, higher frame rates increase processing requirements.

Exposure times may also need to be shorter.

The optimum rate depends on aircraft speed and environment.

High-performance drones balance image quality, frame rate and computing capability.

Camera Resolution

Higher resolution provides more visual detail.

However, VIO does not necessarily require extremely high-resolution imagery.

Processing large images increases computational demand.

Navigation systems may therefore use moderate-resolution cameras optimised for high frame rates and low latency.

A separate high-resolution camera can be used for inspection or mapping.

Navigation-camera specifications should be judged by navigation performance rather than photographic quality.

Dynamic Range

High dynamic range is valuable when a drone moves between bright and dark environments.

For example, an indoor drone may look toward a bright doorway while the surrounding room remains dark.

A camera with limited dynamic range may lose features in either the shadows or highlights.

HDR-capable navigation cameras can improve feature availability.

However, HDR processing must maintain low latency for real-time navigation.

Visual Texture

VIO performs best when surfaces contain identifiable visual texture.

Concrete, brick, machinery and natural terrain often provide many features.

Blank walls, uniform floors or featureless ceilings can be difficult.

Artificial markers can sometimes improve navigation in controlled facilities.

However, a general-purpose autonomous drone should ideally operate without requiring infrastructure modifications.

Multi-sensor navigation provides another solution.

Repetitive Environments

Repeated patterns can create ambiguity.

Warehouse racks, identical corridors and repetitive windows may look similar.

The VIO system can continue estimating local motion, but map-based localisation may struggle to determine the exact location.

Inertial information helps maintain continuity.

LiDAR geometry or known landmarks can provide additional constraints.

Repeated environments demonstrate why localisation confidence should be monitored rather than assumed.

Moving Objects

VIO works best when tracked features belong to stationary objects.

People, vehicles, machinery or moving vegetation can create misleading motion.

Advanced algorithms identify inconsistent features and remove them from the navigation solution.

However, environments containing mostly moving objects remain difficult.

For example, a drone flying above a dense moving crowd should not rely entirely on those people as visual references.

Stable structural features provide stronger navigation anchors.

Water

Water surfaces can be challenging for VIO.

They may contain reflections, changing patterns and limited stable texture.

A downward-facing camera over open water may therefore provide poor visual odometry.

The IMU can maintain short-term movement estimates, but drift will increase.

GNSS, radar, LiDAR or other navigation sources may be more appropriate.

VIO should not be assumed reliable simply because the camera can see the surface.

Snow and Uniform Terrain

Snow-covered landscapes can contain very little visual texture.

Bright conditions may also create exposure challenges.

Similarly, uniform sand or smooth floors can provide few trackable features.

VIO performance can therefore degrade.

Other sensors should support navigation in environments where visual features are expected to be limited.

Smoke and Dust

Smoke and dust reduce image contrast.

They may also create moving patterns that do not correspond with fixed structures.

This can reduce VIO reliability.

Emergency and mining drones operating in these environments may benefit from LiDAR or radar.

Sensor fusion provides redundancy when one sensing modality becomes degraded.

A navigation architecture should be designed around the worst realistic conditions rather than ideal demonstrations.

Vibration

Drone motors and propellers generate vibration.

Excessive vibration can affect both cameras and IMUs.

The IMU may record high-frequency movement that is unrelated to the aircraft’s useful trajectory.

Images may blur.

Mechanical isolation and structural design are therefore important.

However, excessive isolation can allow the sensors to move relative to the airframe.

The mounting system must balance vibration reduction with rigid calibration.

Rolling-Shutter Effects

Rolling-shutter cameras expose image rows sequentially.

When the drone moves rapidly, different parts of the image correspond to slightly different camera poses.

This can distort straight objects.

Modern VIO algorithms can compensate for some rolling-shutter effects.

Nevertheless, global-shutter cameras are often preferred for demanding navigation.

The trade-off depends on cost, weight, sensitivity and system architecture.

Processing Requirements

VIO requires substantial real-time computation.

The system must process images, IMU measurements and optimisation calculations with very low latency.

Modern embedded processors, GPUs and dedicated AI accelerators make this increasingly practical.

However, computational performance also affects power consumption and heat.

Small drones have limited cooling capability.

Navigation computers therefore need to balance performance with size, weight and power.

Edge Computing

VIO is fundamentally an edge-computing application because navigation decisions need to happen onboard.

A drone cannot normally depend on sending every image to a remote cloud server and waiting for the result.

Onboard processing allows navigation to continue even if the communication link is interrupted.

This is particularly important indoors and underground.

Cloud processing may still be used after the mission for mapping and analysis.

Navigation Confidence

Advanced systems can estimate uncertainty in their position.

This is extremely valuable.

Instead of simply outputting coordinates, the navigation system can indicate when confidence is deteriorating.

The flight controller can then respond conservatively.

For example, the drone may slow down, hover, return toward a known area or switch to another sensor.

Future autonomous drones are likely to rely increasingly on this type of navigation-health monitoring.

VIO and Flight Controllers

The VIO system typically sends position, velocity or pose estimates to the flight controller.

Platforms based on systems such as PX4 or ArduPilot can support external navigation information when configured appropriately.

The flight controller then uses these estimates for position control.

However, integration requires correct coordinate frames, timing and covariance information.

Poor integration can produce unstable behaviour even when the VIO estimate itself is good.

System-level testing is therefore essential.

VIO and Autopilots

Autopilots traditionally combine GNSS, IMU, barometer and magnetometer measurements.

VIO provides an additional navigation source.

The autopilot may transition between GNSS and VIO depending on availability.

More advanced architectures can fuse both simultaneously.

The transition should be smooth.

Sudden coordinate jumps can cause undesirable aircraft movement.

Professional integration therefore requires careful reference-frame management.

VIO and LiDAR

Combining VIO with LiDAR provides strong complementary capabilities.

VIO tracks visual features and inertial movement.

LiDAR directly measures surrounding geometry.

In low light, LiDAR can continue operating while visible cameras struggle.

In environments where LiDAR geometry is repetitive, visual texture may provide additional information.

Multi-sensor fusion can therefore create a more resilient navigation system than either technology alone.

VIO and Depth Cameras

Depth cameras provide distance information alongside imagery.

Stereo cameras, structured-light sensors and time-of-flight cameras can all contribute.

Depth information can improve obstacle awareness and mapping.

However, different depth technologies have different range and lighting limitations.

Outdoor sunlight can affect some active infrared systems.

The sensor should therefore be selected for the intended environment.

VIO and Radar

Radar is increasingly being explored for drone navigation.

It can operate in darkness and may perform better than optical cameras in some adverse conditions.

Combining radar with VIO could provide greater resilience.

The camera provides rich visual features, while radar provides range and velocity information.

This type of sensor diversity may become increasingly important for autonomous drones expected to operate in challenging weather or visibility.

VIO and Barometers

Barometers estimate altitude from atmospheric pressure.

They provide useful relative vertical information.

However, indoor pressure can change because of ventilation or weather.

Propeller airflow may also influence measurements.

Barometric altitude can therefore complement VIO but should not be treated as a perfect vertical reference.

Sensor fusion can determine how much confidence to place in each source.

VIO and Magnetometers

Magnetometers provide heading information relative to Earth’s magnetic field.

However, buildings and industrial environments contain metal and electrical equipment that can distort magnetic measurements.

VIO provides relative heading changes without depending on magnetic fields.

Combining both can be useful when magnetic information is reliable.

Navigation software should detect and reject obviously inconsistent magnetic measurements.

VIO and Ultra-Wideband

Ultra-Wideband, or UWB, positioning systems can provide absolute reference points indoors.

Fixed anchors are installed around a facility.

The drone measures its relationship to these anchors.

VIO provides smooth local movement while UWB limits long-term drift.

This can create highly reliable indoor navigation.

The disadvantage is that infrastructure needs to be installed.

VIO alone is more flexible for unknown environments.

VIO and Visual Markers

Facilities can install visual markers such as fiducial tags at known locations.

The drone recognises these markers and corrects its position.

This is a relatively inexpensive way to constrain VIO drift in controlled environments.

However, it requires infrastructure and clear visibility.

Markers can become obscured or damaged.

They are therefore most useful in warehouses and factories where the environment can be managed.

VIO and SLAM Maps

A VIO system can provide the short-term trajectory used by a SLAM system.

The SLAM layer builds a persistent map.

When the drone revisits a known location, the system can recognise it and correct accumulated drift.

This creates a powerful navigation architecture.

VIO handles immediate movement while SLAM provides longer-term spatial consistency.

For autonomous indoor drones, the two technologies are increasingly used together.

Loop Closure

Loop closure occurs when the system recognises a previously visited location.

Suppose the drone flies around a warehouse and returns to its starting area.

Its VIO estimate may have accumulated some drift.

The mapping system recognises known visual features and determines that the current position should match the earlier location.

It can then optimise the trajectory.

Loop closure is one of the major differences between simple odometry and full SLAM.

Relocalisation

A drone may temporarily lose visual tracking.

When useful features return, the system needs to determine where it is.

If a previously created map is available, it may recognise the environment and relocalise.

This is valuable after rapid motion or temporary darkness.

However, successful relocalisation depends on having enough recognisable information.

Safety systems should not assume recovery will always occur.

Map-Based Navigation

Once an environment has been mapped, a drone can use the map for repeated autonomous missions.

This is useful in warehouses, factories and industrial facilities.

The drone can follow inspection routes without GNSS.

VIO provides continuous local motion estimation while map matching corrects drift.

However, environments can change.

New equipment, pallets or construction work may make an old map partially obsolete.

Autonomous systems need to distinguish permanent map features from temporary changes.

Drone-in-a-Box Operations

VIO could become an important technology for indoor and GNSS-challenged Drone-in-a-Box systems.

A drone could launch automatically, navigate through a facility, collect inspection data and return to its station.

Repeat missions could monitor industrial assets or warehouses.

However, reliable autonomous docking requires highly accurate local positioning.

VIO may therefore be combined with visual markers, UWB or precision landing systems near the dock.

Inspection Repeatability

Repeatable navigation is particularly valuable for inspection.

If a drone follows approximately the same route every week, images can be compared more effectively.

VIO and map-based localisation can support this repeatability.

However, relative navigation accuracy should be verified.

A drone returning to approximately the same location does not necessarily reproduce the exact camera viewpoint.

Dedicated visual alignment may be required for precise inspection comparisons.

Mapping with VIO

VIO trajectories can georeference camera or LiDAR measurements within a local coordinate frame.

This supports three-dimensional mapping in GNSS-denied environments.

However, accumulated trajectory drift affects the map.

SLAM optimisation and control points can improve consistency.

If global coordinates are required, the local model needs to be aligned with surveyed references or GNSS observations.

Photogrammetry and VIO

VIO can provide approximate camera positions for photogrammetric processing.

This can improve initial alignment.

However, photogrammetry normally performs its own bundle adjustment using image observations.

For survey-grade outputs, independent control remains important.

The VIO trajectory should therefore be treated as useful navigation information rather than automatically as a survey-quality reference.

Digital Twins

VIO-enabled drones can repeatedly navigate through industrial facilities and collect data for digital twins.

The aircraft may carry RGB, thermal, LiDAR or other inspection sensors.

The navigation system allows measurements to be associated with locations.

This creates opportunities for automated facility monitoring.

However, the accuracy of the digital twin depends on the mapping and georeferencing system, not VIO alone.

AI and VIO

AI can improve visual feature selection, scene understanding and dynamic-object filtering.

Neural networks may identify which parts of an image are likely to provide stable navigation information.

For example, the system can avoid relying on moving people or vehicles.

AI may also help recognise previously visited locations.

However, AI should complement robust geometric navigation rather than remove the need for physical validation.

Semantic Navigation

Future drones may understand not only geometry but also the meaning of objects.

A system might recognise a doorway, corridor, machine or staircase.

This semantic information can support navigation decisions.

VIO provides the motion estimate while AI provides environmental understanding.

The combination could make autonomous drones significantly more capable inside complex facilities.

However, object recognition errors should not directly override fundamental flight-safety constraints.

Navigation Resilience

One of the strongest reasons to deploy VIO is not necessarily to eliminate GNSS but to reduce dependence on any single navigation source.

A resilient drone may combine GNSS, VIO, LiDAR, optical flow, radar, barometric altitude and other measurements.

Each sensor contributes according to current conditions.

If GNSS becomes unreliable, VIO receives greater weighting.

If visual tracking deteriorates, LiDAR or another source may become more important.

This sensor-fusion approach is likely to define the next generation of autonomous drones.

Data Integrity and Cybersecurity

Navigation data is safety-critical.

VIO software, calibration parameters and sensor data therefore need appropriate protection.

Industrial and public-safety drones may require secure software updates, authenticated firmware and protected communications.

However, VIO also has an advantage because much of its navigation information is generated locally onboard.

It does not inherently depend on an external positioning signal.

This can increase resilience, although it does not eliminate other navigation failure modes.

Testing VIO Systems

VIO should be tested in environments representative of actual operations.

A system demonstrated in a brightly lit office may behave differently in a dark tunnel.

Testing should include changes in lighting, feature density, aircraft speed and vibration.

Long trajectories are important because drift may not become obvious during short demonstrations.

Transitions between GNSS and VIO should also be tested if hybrid navigation is expected.

Measuring VIO Performance

Important performance characteristics include positional drift, orientation drift, latency, update rate and recovery from temporary tracking loss.

Performance should be measured over realistic trajectories.

Simply showing that a drone can hover indoors is not sufficient evidence of long-distance navigation accuracy.

Users should ask how the system performs after travelling hundreds of metres or operating for extended periods without absolute position corrections.

Choosing a VIO System for a Drone

Selecting VIO should begin with the operating environment.

Indoor inspection requires different capabilities from high-speed outdoor autonomous flight.

Important considerations include camera configuration, field of view, shutter type, frame rate, low-light sensitivity, IMU quality, synchronisation, calibration, onboard computing, latency, software integration and drift performance.

The ability to integrate other navigation sources is equally important.

A VIO system that works only in isolation may be less useful than one designed for broader sensor fusion.

Benefits of Visual-Inertial Odometry

VIO can significantly expand where drones operate.

It enables relative navigation without continuous GNSS, requires relatively compact sensors and can operate using hardware already present on many modern aircraft.

It can support stable indoor flight, autonomous inspection and navigation through complex environments.

When integrated with SLAM, it can also support mapping and repeated autonomous missions.

The technology is particularly attractive because cameras provide rich information while IMUs are small, lightweight and inexpensive compared with many alternative navigation sensors.

Limitations of Visual-Inertial Odometry

VIO also has important limitations.

It depends on visual features and can struggle in darkness, smoke, dust, fog, featureless environments and scenes dominated by moving objects.

Camera blur can reduce performance.

Incorrect calibration or timing can introduce errors.

Most importantly, VIO accumulates drift because it estimates relative movement rather than continuously receiving an absolute global position.

A successful VIO deployment therefore depends on understanding these limitations and integrating additional navigation sources where necessary.

The Future of VIO for Drones

Visual-Inertial Odometry is likely to become a fundamental component of increasingly autonomous drone systems.

Improvements in cameras, IMUs and onboard processors will allow sophisticated navigation algorithms to operate on smaller aircraft.

AI will improve feature tracking, scene understanding and relocalisation.

VIO will increasingly be integrated with LiDAR, radar and other sensors.

Rather than switching between individual navigation technologies, future drones are likely to maintain a continuously fused navigation solution that evaluates the reliability of every available sensor.

A drone could operate outdoors using GNSS, transition beneath a structure using VIO, enter a dark tunnel using LiDAR-inertial navigation and then recover its global position when satellite reception becomes available again.

This type of navigation resilience will be important for industrial inspection, autonomous logistics, underground operations, emergency response, infrastructure inspection, robotics and Drone-in-a-Box deployments.

A future workflow could operate as:

mission planning → GNSS-supported outdoor navigation → automatic detection of degraded satellite positioning → VIO assumes greater navigation responsibility → camera and IMU continuously estimate motion → LiDAR or depth sensors provide additional geometric constraints → SLAM builds and optimises the local map → AI monitors navigation confidence and environmental conditions → autonomous mission continues → known landmarks or GNSS provide absolute position correction → inspection or mapping data is georeferenced → professional review and analysis.

Conclusion

Visual-Inertial Odometry is one of the key technologies enabling drones to move beyond dependence on satellite navigation.

By combining cameras with inertial measurements, VIO allows an aircraft to estimate its position, velocity and orientation from its movement through the surrounding environment.

This makes it particularly valuable for indoor environments, tunnels, warehouses, mines, industrial facilities, infrastructure inspection, construction and other GNSS-denied or GNSS-degraded locations.

Its greatest strength is the complementary relationship between vision and inertial sensing. Cameras help correct the drift associated with inertial navigation, while the IMU provides rapid motion information when visual measurements alone are insufficient.

However, VIO is not an absolute positioning technology and should not be treated as universally reliable. Visual texture, lighting, motion blur, vibration, calibration, moving objects and environmental conditions can all influence performance, while positional drift can accumulate during long missions.

The strongest drone navigation systems therefore use VIO as part of a broader architecture combining visual sensing, inertial navigation, GNSS, SLAM, LiDAR, depth sensing and other positioning technologies according to the operating environment.

As drones become more autonomous and begin operating routinely inside buildings, beneath infrastructure and underground, Visual-Inertial Odometry is likely to become one of the foundational technologies that allows them to understand not only where they are moving, but how they can continue navigating when conventional satellite positioning is no longer available.