Skip to content

Latest commit

 

History

History
78 lines (63 loc) · 4.17 KB

analysis.md

File metadata and controls

78 lines (63 loc) · 4.17 KB
Raw

Analysis

Overview

  • matching statistics shown during execution of tool
  • export of various data by setting corresponding parameters in config-file
    • data is exported to .txt files that are then read by python-scripts
    • set export path in config-file
    • adjust import paths at the beginning of python-scripts
  • analysis scripts in /analysis

1. Analysis of Geometric Alignment

  • the script is on PATH of the docker container; you just need to specify where you saved the output data:
  plot_traj_matching.py /path/to/output/traj_matching
  • visualization of UTM-projection, Umeyama transformation and Rubber-Sheet transformation
  • execute script plot_traj_matching.py in /analysis
  • produces graphs shown in alignment
  • calculation of deviation between trajectories based on distance between point and line segment
  • distance from point $\vec{p}$ on target trajectory to segment of points $\vec{a}$ and $\vec{b}$ on master:
$$d = \begin{cases} \|\vec{p} - \vec{a}\| & \quad u < 0. \\\ \|\vec{p} - \vec{b}\| & \quad u > 1. \\\ \frac{|(\vec{b} - \vec{a}) \times (\vec{p} - \vec{a})|}{\|\vec{b} - \vec{a}\|} & \quad 0 \leq u \leq 1. \end{cases}$$

where

$$u = \frac{(\vec{b} - \vec{a}) \cdot (\vec{p} - \vec{a})}{\|\vec{b} - \vec{a}\|}$$
  • minimum deviation calculated by minimum distance of $\vec{p}$ to any segment of subsequent points
  • note that this produces falsified results near intersections of the trajectories

2. Analysis of Matching/Conflation

  • the script is on PATH of the docker container; you just need to specify where you saved the output data:
  plot_matching.py /path/to/output/matching

  • command window

    • matching statistics shown in command window during execution indicate
    • geometric similarity measures calculated as in matching algorithm.
    • mean values of $\beta$, $l$, $d$, $\bar{S}$ and $score$

  • script plot_matching.py in /analysis

    • (un)comment the desired modules in the main function

    • calculation of similarity measures and visualization

    • visualization of matching results (match polylines, linkages, original lanelets, updated lanelets)

    • calculation of precision and recall

      precision = \frac{True~Positive}{True~Positive + False~Positive} \\[5pt]%
recall = \frac{True~Positive}{True~Positive + False~Negative}
      Prediction/Reality True False
      Positive Algorithm correctly identifies the corresponding OSM polyline for a given reference polyline. Algorithm predicts a match but the prediction is wrong (either wrong OSM polyline or no match at all).
      Negative Algorithm correctly identifies that there is no match for a given reference polyline. Algorithm does not identify a match for a given reference polyline when in fact there is one.

    • match considered True Positive if $score$ > 0.8, otherwise False Positive

    • identification of false negatives by manual investigation (use module identify_fn() inside plot_matching.py)

3. Analysis of Georeferencing

  • visualize lanelet map and GNSS trajectory in comparison to orthophoto
  • orthophotos from Bavaria available on OPENDATA
  • orthophotos are rectified in UTM32 (EPSG:25832) coordinates and have a size of 1km x 1km each
  • script plot_georef.py in /analysis
    • download photos from desired area on OPENDATA
    • adjust path of image and set photo extent (boundaries of photo in UTM-coordinates)