Volvo Cars has developed a map attributes representation as a graph in Neo4j. By including real time car data, they are able to collect insights to learn on possible accident causes based on road infrastructure.
6. Graph Summit, Johan Amoruso
2023.03.16
AI AWARE
Friction Speed Traffic Road-works Weather Map Accidents
Accident Risk Alert
7. 2023.03.16 Graph Summit, Johan Amoruso 7
A
B
C
D
Road infrastructure
Segments A, B and C:
Roads
Junction D:
- A connects to B
- C connects to A
- …
Mapping geometry
8. 8
Road infrastructure
Segments A, B and C:
Driving directions
Junction D:
- A- connects to B-
- C- connects to A+
- …
B+
C-
B-
C+
D
A+
A-
Mapping geometry
2023.03.16 Graph Summit, Johan Amoruso
13. Accident risk exploration
13
Connected feature extraction
Extracting features from the graph to be usable
in a ML pipeline
Friction
Traffic
Weather
Daylight Accident risk
2023.03.16 Graph Summit, Johan Amoruso
14. Connected Feature Extraction
14
Generated in two main ways:
1) Running local graph queries
Querying specific nearby infrastructure:
- Roundabouts
- High curvatures
- Decreased speed limit
2) Running global graph algorithms
Learn about the general structure:
- Community detection
- Graph embeddings
Nearby roundabouts
Curvatures
2023.03.16 Graph Summit, Johan Amoruso
15. 15
Data from 2022-09-15
to 2022-10-31
# Aggregations: 10,000
# Accidents: 183
Local graph queries
Accident
Normal
Condition
2023.03.16 Graph Summit, Johan Amoruso
17. Global graph algorithms
17
GraphSAGE embedding
Features:
- Maximum Speed Limit
- Max number of lanes
- Min number of lanes
- Functional class
- Roundabout
n ~ 100,000
Accident
Normal
Condition
2023.03.16 Graph Summit, Johan Amoruso
18. Challenges
18
- Graph design. So many options!
- Multiple graphs... Graph projections.
- Sparse data: Not all attributes are available on every node.
- Currently not considering the time aspect.
2023.03.16 Graph Summit, Johan Amoruso