Giving Machines the Power to See More on Our Roads

Much of the pressure for the safety of automated vehicles lies on the means of transport themselves.

New vehicles are equipped with safety features such as ‘lane keep assist’, ‘lane centring’ and ‘automatic steering functions’, but what good are they if the road infrastructure that they operate on are not designed with these features in mind?

 

If road operators optimised pavement markings and road design, could this help reduce road trauma on Australian and New Zealand (ANZ) roads?

 

This is the question WSP and Austroads set to answer in their research into the automated steering functions in current vehicle models and detection of lane positioning. Here, Julien Marr (JM), our Intelligent Transport Lead for Victoria shares some detail into the research and how machine vision works.

 

How does machine vision work and how is it regulated in ANZ?

JM: Machine vision looks to construct lines from what it observes over time using basic algorithms to determine where a line is. This relative position is then used to provide information, often to a sensor suite, to support safe functions such as lane keep assist, lane centring and automatic steering functions.

 

We use a variety of standards and guidelines but there has recently been a welcomed movement towards harmonisation locally. A lot of machine vision quality issues stem from poor maintenance and implementation of standards and guidelines.

 

What methodologies were used to undertake the current research?

JM: A literature review was conducted to help focus questions for stakeholders and later inform field tests that were conducted in February. Stakeholder engagement was key to understanding the issues of importance, the potential impact of policy changes.

 

Initially, we conducted a pilot test to check that our testing methodology was sound and would deliver appropriate results.

 

Field testing was then undertaken to test potential policy or design changes for pavement markings and the potential impact this might have on vehicles. On-road scenarios helped gather large volumes of information to assist with disaggregated analysis picking up patterns that we were not even aware of. It also helped us test some complex designs and specific conditions (e.g. different pavement types) that are difficult to replicate in a controlled environment. The off-road scenarios were to provide a stronger evidence base for relatively high-cost infrastructure decisions such as pavement marking contrast and thickness.

 

Finally, a cost analysis was developed to try and understand the cost implications of our proposed standards in Australia and New Zealand to improve their pavement markings to better support connected and automated vehicle functions.

 

What were the findings and what are some possibilities for the future?

JM: By implementing some simple design and a moderate increase in investment in targeted pavement markings, substantial road safety benefits of lane keep assist functions are attainable.

 

Julien Marr will be co-presenting with Austroads at the 2019 Australian Institute of Traffic Planning and Management National Conference on Pavement Markings for Machine Vision on Thursday 1 August at 3.00pm.

 

To stay abreast of our latest news, publications, videos and posts, please follow us on LinkedIn.