Program Title
Feature Extraction
How might we better
Identify and extract features of interest from video and pointcloud data?
Problem Story
There are thousands of road assets spread across the 18000kms of roads in WA. These include signs, barriers, line marking, pavement sealed widths, shoulder widths etc.
With thousands of different assets to keep track of and the need to know not only where they are but also details like their distance from the road edge.
If we were able to automate the capture of asset data through the extraction of features from some of our visual data sets (video and laser point cloud data), this would allow us to update or validate our asset databases more cheaply and quickly.
The challenge for the Ministry of Data is, can we automate the detection and extraction of the relevant features of our network from these data sources.
With thousands of different assets to keep track of and the need to know not only where they are but also details like their distance from the road edge.
If we were able to automate the capture of asset data through the extraction of features from some of our visual data sets (video and laser point cloud data), this would allow us to update or validate our asset databases more cheaply and quickly.
The challenge for the Ministry of Data is, can we automate the detection and extraction of the relevant features of our network from these data sources.
Scale / Impact
HIGH
Video data already captured
Laser scan point cloud data available
Potential hazard to road users if assets misidentified or not recorded
Cost to manually capture asset data
Data Description
Main Roads
Asset data
Video surveys
Mobile Laser scan point cloud data
External
Satellite data?
Asset data
Video surveys
Mobile Laser scan point cloud data
External
Satellite data?
4 key outcomes / constraints / statements
- 18,000km of rural road network
- Manual data capture and review process
- Thousands of asset classes
- Complex interelational data points required