ABSTRACT
Road infrastructure is one of the pillars of modern societies: it allows efficient transport of people and goods where other transport modalities could not compete in terms of cost efficiency. However, road asphalt tends to deteriorate over time, with use, and due to atmospheric and environmental phenomena. It is precisely the capillarity of road infrastructure, its greatest value, which also makes it very difficult and expensive to monitor and maintain, thus making road monitoring an important part of structural health monitoring (SHM). To date, in most countries, the detection of damage to the road surface is done manually, with specialized operators in the field operating expensive equipment, greatly limiting the effectiveness of monitoring by maintenance bodies. The recent breakthroughs in computer vision, neural networks, and artificial intelligence make low-cost monitoring of infrastructure possible and scalable, reducing the reliance on monitoring performed by more expensive and invasive methods that require human intervention to just a small number of critical cases. In this chapter, we overview and discuss a number of methods for pothole and crack detection on the road surface, focusing on the recent achievements in semantic segmentation of road images to identify both potholes and cracks using deep learning techniques. In addition, we present some of the devices for data acquisition currently available on the market, focusing in particular on depth cameras, and we introduce the most important open datasets and benchmarks currently available in the literature. Finally, we discuss future research directions regarding the use of RGB-D (Red Green Blue-Depth) technology and depth cameras as an additional source of information to be provided to the neural network during training or to set up self-supervised learning pipelines.
