ov_eval namespace

Evaluation and recording utilities.

This project contains the key evaluation and research scripts for localization methods. These come from the necessity of trying to quantify the accuracy of the estimated trajectory while also allowing for the comparision to other methods. Please see the following documentation pages for details:

• Filter Evaluation Metrics — Definitions of different metrics for estimator accuracy.
• Filter Error Evaluation Methods — Error evaluation methods for evaluating system performance.
• Filter Timing Analysis — Timing of estimator components and complexity.

The key methods that we have included are as follows:

• Absolute trajectory error
• Relative pose error (for varying segment lengths)
• Pose to text file recorder
• Timing of system components

The absolute and relative error scripts have been implemented in C++ to allow for fast computation on multiple runs. We recommend that people look at the rpg_trajectory_evaluation toolbox provided by Zhang and Scaramuzza. For a background we recommend reading their A Tutorial on Quantitative Trajectory Evaluation for Visual(-Inertial) Odometry [47] and its use in A Benchmark Comparison of Monocular Visual-Inertial Odometry Algorithms for Flying Robots [10]

Classes

class AlignTrajectory

Class that given two trajectories, will align the two.

class AlignUtils

Helper functions for the trajectory alignment class.

class Loader

Has helper functions to load text files from disk and process them.

class Recorder

This class takes in published poses and writes them to file.

class ResultSimulation

A single simulation run (the full state not just pose).

class ResultTrajectory

A single run for a given dataset.

struct Statistics

Statistics object for a given set scalar time series values.