Using crowdsourced data to improve road management

This project will demonstrate ability to use historic crowdsourced data to create value in road management process by prototyping tools to proactively support road network management.

The traditional model of collecting road asset information through once-a-year drive using camera and/or other sensors to provide asset information is useful in extracting required information through manual means.

Although there are machine learning tools to extract certain attributes from the image/points cloud, so far, they have proven to be unreliable. Moreover, extracted information could be several months old and therefore currency of the information extracted may not be useful.

Automated vehicle and associated technology have now evolved to map the road network in near real-time through edge computing. The produced maps (also known as HD maps or AV maps) of the environment are critical in achieving high level of automation.

Enabled by fast 4G/5G communication, car manufacturers and automotive equipment suppliers are collecting this data as well as vehicle dynamics data at large scale to create a more accurate, more dynamic, representation of the road environment.

With increasing numbers of sensors being deployed on our roads in new vehicles, there is a significant increase in crowdsourced road network data being collected.

This project will access one or more commercially available crowdsourced data and:

• Verify the accuracy and the currency of the data
• Build tools to address several road operator use cases using this data, including for maintenance and safety

Participants

• Department of Transport and Main Roads (Queensland)
• Queensland University of Technology
• CARRS-Q

Project background

The Queensland Department of Transport and Main Roads (TMR), and the Queensland University of Technology (QUT), are delivering the project to help understand how new vehicles fitted with sensing technologies and communication capabilities can provide a substantial benefit in road management.

Research need

There is a significant body of research that has investigated the use of computer vision (or machine learning using images) to read signs and/or identify assets in the road corridor. The projects highlighted the limitations of the technology where a single set of images is used. Accuracy and currency of the data is not at a level required to support the data needs for effective road management.

Probing road infrastructure using edge-computed crowdsourced data

One approach used to address the data shortfall is to collect more images over the same area. With more images, both accuracy and completeness improve. Several actors have used this approach to crowdsource events on a map (such as waze), and even generate sufficiently accurate data for the automated driving (Tesla).

Among those actors, for example, Mobileye generates edge computed computer vision data and collects the generated data through crowdsourcing to create accurate and dynamic maps. Mobileye is in the forefront of this initiative. As of August 2021, Mobileye claims to be mapping some 7-8 million kilometres of road each day with about 10cm accuracy. The company further claims that it will have the ability to map one billion kilometres of global roads each day by 2025.

This level of scalability has been made possible through Mobileye’s agreements with several passenger car manufactures around the world. There are currently millions of cars on the road (including in Australia) with Mobileye technology onboard. These vehicles contribute to crowdsourced highly-accurate maps of the road network each time they drive.

Mobileye data is only included in the dataset once the feature has been identified and classified following several runs. This crowdsourced method of data collection provides continuous data improvements and the potential for identifying changes in road or asset conditions in near real time.

Project intent

This project aims to address some needs of road authorities related to road management, by using crowdsourced information. Three needs have been identified as being critical and potentially providing high value:

1. Automated safety risk assessment tool of the road infrastructure
2. Automated road asset monitoring, including asset degradation
3. Automated intersection map including vehicle trajectory

The project will not create its own dataset however will source edge computed computer vision crowd sourced (and potentially other vehicle dynamics data) data at defined interval and build tools for the above three tasks.

It is proposed to develop a proof of concept that will cover 100 kilometres of varied road environment. If needed, edge computing equipment will be sourced and installed in vehicles to allow the user to control the time and location variables and ensure data is available for the desired location at the desired time.

Project objectives

The project objective is to assess the viability of using data that has been crowdsourced and processed through automated tools addressing following three identified road operator needs:

1. Automated safety risk assessment tool of the road infrastructure: the Australian Road Assessment Program (AusRAP) defines a method to evaluate road segments. This automated assessment tool will automatically produce an assessment of each segment of road using the AusRAP methodology and edge computed computer vision crowdsourced data. This type of tool would allow any road to be assessed at any time, resulting in an improved understanding of road safety and enabling improved investment decisions.
2. Automated road asset monitoring, including road asset degradation: the road asset maintenance planning tool will monitor asset degradation and provide recommendations for when maintenance activity is required for the asset. This may include the road surface, signage, poles or other assets. The assets to be included will be determined based on data availability (& perceived quality) in the 100-kilometre road segments chosen by the project in consultation with TMR’s asset management team. This type of tool has the potential to improve maintenance efficiency, reduce urgent repairs and improve road safety.
3. Automated intersection map including vehicle trajectory, the intersection road topology (MAPEM) creation tool will create detailed maps of intersections in accordance with international C-ITS standards. Existing processes for this task are labour-intensive, and quality is variable due to the manual nature of the process. This type of tool would allow intersection details to be mapped quickly, efficiently and repeatably across entire road network with minimal human input.

These tools will be created at a proof-of-concept level, with further investment necessary to create robust tools that could be used on a commercial or ongoing basis.

Please note …

This page will be a living record of this project. As it matures, hits milestones, etc., we’ll continue to add information, links, images, interviews and more. Watch this space!