Library overview: mrpt-base

Back to list of libraries

mrpt-base

---

This is the most fundamental library in MRPT, since it provides a vast amount of utilities and OS-abstraction classes upon which the rest of MRPT is built. Here resides critical functionality such as mathematics, linear algebra, serialization, smart pointers and multi-threading.

This library comprises classes in a number of namespaces, briefly described below (click on the namespaces names to see the complete list of its classes):

mrpt::poses

A comprehensive collection of geometry-related classes to represent all kind of 2D and 3D geomtry transformations in different formats (Euler angles, rotation matrices, quaternions), as well as networks of pose constrains (as used typically in SLAM problems).

There are also implemented representations for probability distributions over all of these transformations, in a generic way that allow mono and multi-modal Gaussians and particle-based representations. See mrpt::poses for the complete list of classes here.

mrpt::utils

• RTTI (RunTime Type Information): A cross-platform, compiler-independent RTTI system is built around the base class mrpt::utils::CObject.

• Smart pointers: Based on the STLplus library, any class CFoo inheriting from CObject, automatically has an associated smart pointer class CFooPtr. MRPT implements advanced smart pointers capable of multi-thread safe usage and smart pointer typecasting with runtime check for correct castings.

• Image handling: The class mrpt::utils::CImage represents a wrapper around OpenCV IplImage's, plus extra functionality such as on-the-fly loading of images stored in disk upon first usage. The internal IplImage is always available so OpenCV's functions can be still used to operate on MRPT images.

• Serialization/Persistence: Object serialization in a simple but powerful (including versioning) format is supported by dozens of MRPT classes, all based on mrpt::utils::CSerializable.

• Streams: Stream classes (see the base mrpt::utils::CStream) allow serialization of MRPT objects. There are classes for tranparent GZ-compressed files, sockets, serial ports, etc.

• XML-based databases: Simple databases can be mantained, loaded and saved to files with mrpt::utils::CSimpleDatabase.

• Name-based argument passing: See the structure mrpt::utils::TParameters

• Configuration files: There is one base virtual class (mrpt::utils::CConfigFileBase) which can be used to read/write configuration files (including basic types, vectors, matrices,...) from any "configuration source" transparently (an actual configuration file, a text block created on the fly, etc.).

mrpt::math

MRPT defines a number of generic math containers, which are:

• Matrices: Dynamic-size matrices (see mrpt::math::CMatrixDouble) and compile-time fixed-size matrices (see mrpt::math::CMatrixFixedNumeric, mrpt::math::CMatrixDouble33, etc.).
• Matrix views: Proxy classes that allow operating on the transpose, a part of, or the diagonal of another matrix as if it was a plain matrix object. See mrpt::math::CMatrixView.
• Vectors: Dynamic-size vectors. See mrpt::vector_double, which inherits from a standard STL vector<double>.
• Arrays: Fixed-size vectors, just like plain C arrays but with support for STL-like iterators and much more. See mrpt::math::CArrayDouble<>.

These contaners have a number of characteristics in common (STL-like iterators and typedefs, etc.) and can be mixed in operations without problems. For example, matrices of any kind can be operated together, a vector can be added to an array, or the results of a matrix operation stored in a matrix view.

Notice that fixed-size contaners should be preferred where possible, since they allow more compile-time optimizations.

Apart from the containers, this namespace contains much more functionality:

• A templatized RANSAC algorithm.
• Probability distribution functions.
• Statistics: mean, covariance, covariance of weighted samples, etc... from sets of data.
• A huge amount of geometry-related functions: Lines (mrpt::math::TLine3D), planes (mrpt::math::TPlane3D), segments, polygons, intersections between them, etc.

mrpt::bayes

Here there are two main family of algorithms:

• Kalman filters: A generic, templatized Kalman filter implementation (includes EKF,IEKF and in the future, UKF), which only requires from the programmer to provide the system models and (optinally) the Jacobians. See mrpt::bayes::CKalmanFilterCapable
• Particle filters: A set of helper classes and functions to perform particle filtering. In this case the algorithms are not as generic as in Kalman filtering, but the classes serve to organize and unify the interface of different PF algorithms in MRPT. See mrpt::bayes::CParticleFilter.

mrpt::synch

This namespace includes threading tools such as critical sections, semaphores or utilities such as the template mrpt::synch::CThreadSafeVariable that converts any variable into a pair variable-critical section.

mrpt::scan_matching

Under this name we find functions in charge of solving the optimization problem of aligning a set of correspondences, both in 2D and in 3D. Note that this does not includes the iterative ICP algorithm, included in another library (mrpt-slam).

mrpt::system

Here can be found functions for filesystem managing, watching directories, creating and handling threads in an OS-independent way, etc.

mrpt::compress

GZip compression methods can be found in this namespace.