Skip to content

Glossary of Terms

Computer vision and perception is full of borrowed terms and phrases. Many of these aren't consistently used from context to context. This is big trouble for someone trying to explain how or what they want to accomplish with perception.

At Tangram Vision, we aim to avoid this kind of ambiguity whenever possible. The terms listed below have a specific definition when used in the platform and throughout this documentation. We hope that this helps conceptualize what the platform accomplishes; in the best case, it may even help you articulate your own perception goals.

A statistical optimization process that aims to solve component model, constraints, and object space collectively. The Tangram Vision Platform's multi-sensor calibration system is called TVCal.
Any hardware that emits observations, or that is part of a spatial or temporal relationship with another component in a Plex.
A spatial, temporal, or root relation between any two components.
The square of the standard deviation of a parameter. Also known as variance-covariance. This value is used as an indication of uncertainty in a Plex.
A piece of hardware that may or may not comprise one or more components. This term is avoided in the Tangram Vision Platform due to its ambiguity.
Extrinsics are how two components are spatially oriented relative to one another. Part of a spatial constraint.
The specific model values for a component. The model used differs between and among component types. Part of a component.
A specific datum produced by a component.
Tangram Vision's universal data structure for describing perception systems. It is a graph of components and constraints that fully describe the system being used.
A piece of hardware. This term is not used frequently at Tangram Vision due to its ambiguity.
A series of observations from a component.
The physical counterpart to a Plex. This represents a perception-enabled machine, robot, vehicle, platform, etc., and all the components it utilizes.
An object with strict geometric properties that is used to produce and capture object points in world coordinates. A common example of this is a checkerboard pattern used for camera calibration.