Sensor calibration is the process of determining the parameters that describe how a sensor maps the world into its measurements, and how sensors are positioned relative to one another. Intrinsics capture a sensor's internal properties, extrinsics capture its pose in space, and transforms relate one sensor's frame to another's. Accurate calibration is what lets measurements from different sensors be combined into a single, consistent view of the world.
Sensor calibration is the process of pinning down exactly how a sensor turns the physical world into numbers, and where that sensor sits relative to everything else on the machine. It is usually described with three related ideas. Intrinsics are the internal characteristics of a single sensor, such as a camera's focal length, optical center, and lens distortion, which govern how points in the world project into the image. Extrinsics describe the sensor's pose, meaning its position and orientation in space. Transforms are the mathematical relationships that convert coordinates from one sensor's frame into another's.
These parameters matter because a raw measurement is only meaningful once you know the geometry behind it. Two cameras pointed at the same object will each report it in their own coordinate system, and only calibration lets you say that both are looking at the same point in the real world. When calibration is accurate, data from cameras, lidar, and other sensors can be projected into a common frame and combined. When it drifts or is wrong, the same object appears in inconsistent places across sensors, and any fusion built on top inherits that error.
Key takeaways
Calibration determines how a sensor maps the world into measurements and how sensors are positioned relative to each other.
Intrinsics describe a sensor's internal properties, extrinsics describe its pose in space, and transforms relate one sensor's frame to another's.
Accurate calibration is what lets multiple sensors be projected into a common frame and fused into a consistent view.
What sensor calibration provides
The three kinds of calibration parameters and what each one describes.
The three kinds of calibration parameters and what each one describes.
Parameter
What it describes
Intrinsics
A single sensor's internal properties, such as focal length, optical center, and lens distortion
Extrinsics
A sensor's position and orientation in space, relative to a reference frame
Transforms
The mapping that converts coordinates from one sensor's frame into another's
How it works
Calibration is typically found by observing known references and solving for the parameters that best explain what each sensor measured. For intrinsics, this often means capturing a target with a known pattern, such as a checkerboard, and computing the internal model that maps the pattern's true geometry to its appearance in the sensor. For extrinsics and transforms, it means observing the same features across sensors and solving for the relative poses that make their observations agree. Because mounts can shift and components can drift with temperature and use, calibration is not necessarily a one-time step, and systems may need to check or refine it over their operating life.
Why it matters
Calibration is the quiet foundation beneath all multi-sensor perception, and its accuracy sets a ceiling on how well fusion, 3D detection, and mapping can work. It matters to anyone building or curating data for physical AI, because errors in calibration masquerade as errors everywhere else, subtly corrupting labels and estimates in ways that are hard to trace back to their source. Getting calibration right, and keeping it right over time, is what allows a machine's many sensors to speak in a common language about a shared world.
Frequently asked questions
What is the difference between intrinsics and extrinsics?
Intrinsics describe a sensor's own internal properties, such as a camera's focal length and lens distortion, which determine how the world projects into its measurements. Extrinsics describe where the sensor is, meaning its position and orientation relative to a reference frame.
What are transforms used for?
Transforms convert coordinates from one sensor's frame of reference into another's. They are what let a point detected by a camera be matched to the same point detected by lidar, which is essential for fusing multiple sensors.
Does calibration need to be redone over time?
Often, yes. Physical mounts can shift, and components can drift with temperature and wear, so calibration can degrade during operation. Many systems periodically check or refine their calibration rather than assuming a single factory setup holds forever.
How does calibration relate to synchronization?
Calibration aligns sensors in space, while synchronization aligns them in time. Combining multiple sensors accurately usually requires both, so that measurements correspond to the same place and the same moment.