June 8, 2018 | Written by: Dan Cunnington
Categorized: Innovation | Research
Share this post:
Supported by the Hartree Centre’s Innovation Return on Research programme – Emerging Technology and colleagues from the STFC Hartree Centre have been working to contextualise the driving environment, fusing multiple data sources together to allow vehicles to make intelligent decisions. This work was recently presented at the IEEE VTS Vehicular Technology Conference in Porto, Portugal alongside other driver assistance technologies.
The paper outlines a data fusion approach, utilising an agile ontology modelling framework called Controlled English (CE) – developed by Emerging Technology in our recent NIS-ITA research programme. CE provides the capability to unify the information space of the driving environment, by specifying domain models, facts and rules using a restricted subset of natural language. For example, in this project we have represented Paxion’s Mental Workload and Driving model in CE to allow a machine (or in this case a vehicle) to evaluate the model and answer questions in a similar manner to a human. As Paxion outlines, driver performance is affected by cognitive load, which is affected by scene complexity and different scenes contain different features such as road type, road curvature and traffic flow.
To enable a vehicle to reason and make decisions in a similar manner to a human, we apply data sources to the CE model and use the Hudson capability for querying a CE Store. Using publically available Open Street Map (OSM) data for the UK, we have extracted various static road features, including roundabouts, traffic signals, pedestrian crossings, inclines and declines, road type and road curvature. These features are modelled in CE and therefore allow high level natural language questions to be asked, such as “Show me all the roundabouts and curvy roads in Winchester”, “Show me areas of high cognitive load between Hursley and Winchester” and “Which features in Winchester are likely to lead to poor driving performance?”
Here you can see residential roads and road segments with high curvature have been extracted for the city of Winchester as these are features of a highly complex scene that (because of Paxion’s model) we can infer to represent features of high cognitive load and therefore features likely to lead to poor driving performance.
We have also extracted dynamic features of the road network, by applying deep learning techniques to on-board vehicle dash cam’s and Road Side Units (RSU’s) to detect traffic flow and the presence of moving or stationary vehicles surrounding the test vehicle.
Contextualising the driving environment by fusing together multiple data sets as well as representing theoretical models of the driving environment allows us to build an intelligent decision making system for vehicles.
Our paper is due to be published on IEEE Xplore in the coming months.