Skip to content

Content types for AI training

Sensor and IMU data collection: specifications before price

Define channels, units, frames, timestamps and calibration before quoting inertial or synchronized sensor data.

For
Data owners and contributors
Reading time
5 min
Updated

A sensor file has little meaning without units, coordinate frames, sampling information and context. Price the defined measurement package and collection work, not a raw row count that can be inflated by sampling more often.

An IMU dataset may contain acceleration and angular velocity, with orientation supplied by a separate estimation process. State which quantities are measured, estimated or unavailable before discussing volume or value.

Planning worksheet — not market rates or a Fiund offer
ItemDefine before requesting a quote
ChannelsUnits, axes, device range and measured versus estimated
TimeClock domain, interval checks and missing samples
CalibrationProcedure, coefficients and applicable sessions
Quote unitAccepted session or elapsed duration, not unexplained rows

Define each channel and coordinate frame

Record device model, firmware, sensor range, configured sample rate, channel names, units and axis directions. Document the mounting frame and how it relates to any camera, body or object frame. Retain raw values and describe conversions.

For ROS-compatible exports, the sensor_msgs/Imu definition specifies acceleration in m/s² and rotational velocity in rad/s. Follow the chosen format explicitly; do not assume data recorded in g or degrees per second already matches it.

Make time and gaps inspectable

Save timestamps, clock source and time units. Separate device capture time from host arrival time where available. Measure actual sampling intervals and document gaps, resets and resampling rather than describing the configured frequency as measured performance.

For synchronized video or other sensors, identify the common time reference or alignment procedure and its measured limitations. A file with the same session name is not evidence that the streams align accurately.

Describe uncertainty and derived signals

Include calibration procedure, date, coefficients and known operating limits when available. If orientation comes from sensor fusion, name that processing and version; do not label it ground truth without the corresponding reference method.

ROS provides specific covariance conventions for unknown or unavailable estimates. Preserve the convention in a compatible export rather than filling every missing measurement with plausible-looking zeros. Other formats need an equally explicit missing-value definition.

Scope the quote around accepted sessions

Budget setup, access, calibration, recording, synchronization, review and rework as separate activities. Agree whether the licence unit is accepted session, elapsed hour or a defined package. More rows per second do not automatically create more useful independent experience.

Send a pilot with raw logs, channel dictionary, session manifest, processing notes and an example of any known gap. Ask the buyer to confirm parsing and suitability before scaling. Fiund can assess an existing inventory or collection capability; no universal IMU licence rate or guaranteed assignment is offered here.

Sources and further reading

Frequently asked questions

Is orientation always measured directly by an IMU?

No. Describe the sensor outputs and any orientation estimation process separately.

Can I quote by number of rows?

You can define that unit in a proposal, but it needs sampling, duration and coverage context. Resampling does not create new physical observations.

Planning a collection project?

Describe collection capabilities →

Related resources

← Content types for AI training