A biomimicry approach to automation of road feature surveys

Australia's road network is a very significant capital asset created by progressive investment over several decades, and is integral to the social and economic performance of the nation. Physical elements of the road infrastructure are numerous and include surfacing, bridges, lighting, signs and line-markings. Road condition assessment is critical to optimisation of both road performance and public investment strategies. Road condition data are of fundamental importance if a road owner is to demonstrate that best practice road management is being followed in line with investment and budgetary constraints. Road surveys that are able to quickly and efficiently identify features and assess their condition are therefore keystones of an effective road asset management system. Manual visual surveys are subjective and expensive, while digital imagery and off-the-shelf computer hardware and software packages have the potential to carry out image analysis in real time. This in turn has triggered a surge of interest in the possibility of automating road survey systems. Efforts to date have stopped short of producing such survey systems. This research was motivated by a desire to understand why this is so, and to explore the way forward. The thesis therefore starts with a general review of the current condition of road assets in Australia, and the manual and semi-automated road condition survey systems widely used by asset managers. The key problems appear to be the difficulties in producing survey systems for quite different applications and survey conditions. Biomimicry suggests that a way forward might be to identify a generic system design, rather than seek to develop a single do-all survey system. The thesis then examines the suite of image processing and artificial intelligence tools, including expert systems and neural networks, that have become available to help automate such surveys. A generic design approach is proposed to develop automated road feature and condition survey systems, together with design application principles that are also largely inspired by biomimicry. The three principal system components are: 1. Image acquisition, which mimics the work of the eyes. 2. Image processing that parallels the work of the visual cortex. 3. Feature recognition and condition assessment that mimic the cognitive aspects of a human brain, notably its biological neural networks. Analysis of the survey results, if necessary, constitutes a fourth system component. This often involves consideration of context and/or aggregation of information into one or more overall road condition indicators. Expert rule systems are ideally suited to such tasks. A two-stage survey development process is proposed. First, a basic survey system is developed which performs well under ideal survey conditions, with the goal of understanding and establishing the fundamental aspects of the survey system. Next, an extended survey system is developed that can handle the more complex, non-ideal survey environments that must be addressed in an actual survey of a long stretch of rural highway. The extended survey system development process starts by examining the performance problems of a basic survey system applied to the more challenging survey conditions. The system design is illustrated by case studies of guidepost counting and centrelinemarking condition assessment. The satisfactory performance of the basic automated systems for these applications is demonstrated using typical sections of road. An extended system for guidepost counting is then produced and its performance examined on a survey of the Midland Highway in Tasmania. The thesis concludes by discussing how best to develop these automated systems into an operational tool through trials carried out in association with local government and road authorities. Case studies are presented to demonstrate the basic survey system development process applied to guidepost counting and to evaluate centreline-marking condition. The approach to developing an extended survey system is demonstrated to produce an improved guidepost survey system able to deal with changing ambient light conditions. The satisfactory performance of this extended system under non-ideal conditions is confirmed via a 50 km field trial on a section of the national highway in Tasmania.