Skip to content

Locating and Exploring Pedestrian Scenarios

Summary

With Edge you can quickly find any scenarios you need from within the knowledge graph, explore their taxonomic coverage, and prepare them for an upcoming risk assessment or simulation test.

As introduced in the Exploring Heterogeneous Datasets from Across the Globe workflow, the AV Sandbox dataset contains a fusion of structured and unstructured datasets, static video data from accident hotspots, dash-cam videos, media articles, and simulated recreations of risky events to map out the tricky events and phenomena autonomous vehicles face on the road.

Many of these involve pedestrians, but how can we quickly identify and extract them from the map?

Identifying Scenarios of Interest

Let's start by opening up the spatial and taxonomy view: some Open Views in Edge →

s

  1. Use search to locate the "pedestrian_hazard" header node fom the taxonomy view.
  2. Select the successors of this feature node. The scenarios involving pedestrians will now be selected and visibly highlighted in the spatial view.
  3. Explore the geospatial coverage and content of these pedestrian scenarios by zooming and panning) through the spatial view!
How to interact with the scenarios

Each node in the "Example Data Sources - Spatial View" represents an real event which took place on roads across the USA and UK. The black nodes are accident reports from FARS/ STATS19 collisions, deep purple nodes are high-risk scenarios captured on public CCTV or Dashcams, and deep blue nodes represent images and videos captured by the dRISK team. The light blue nodes located around San Francisco contain accident descriptions involving Autonomous Vehicles.

You can select any purple or blue node and hit w to watch the video, select any black incident report node

and use the Inspect Agent to read through all the details recorded, and any light blue node and hit . to read the written AV Collision report.

Read more in Interacting With Your Data.

Creating a Pedestrian Risk Map

  1. Select the successors of the "pedestrian_hazard" node, as we have just done from our preliminary exploration above.
  2. Create a New View - your current selection of scenarios will be added to the new view at the same position they were located on the world map view.
  3. At this point, we can close our original spatial view
  4. Rename the new view to "Pedestrian Risk Map", and turn on the google-maps background

In just a few clicks, we now have a map of real-world scenarios that involve the hazard type of interest to us - pedestrians!

Assessing Taxonomic Coverage

Now we have all our pedestrian scenarios in a dedicated view, let's evolve it from a spatial view to a taxonomic map.

  1. Browse through the "Full Scenario Taxonomy" view and iteratively select the features which are relevant to pedestrian scenarios
  2. Copy & Paste them into our new view called "Pedestrian Risk Map"
  3. Turn off the map background
  4. Force direct the view to cluster the pedestrian scenarios around the feature nodes which they connect to

s

Read, Watch, Explore!

We can now simply watch through the videos attached to these scenarios and explore the pedestrian risk space.

You will observe a range of risk sources:

  • Simulated scenarios
  • Dashcam videos
  • JamCam videos
  • Mocab suit simulations

The following short video covers this full workflow:

By interacting directly with our data we were able to locate, extract, and cluster scenarios of interest within just a few minutes of opening the views in Edge.

Last update: 2024-06-07