Úthiba-felismerés okostelefonnal, konvolúciós neurális hálózatok segítségével

Abstract

The application of deep neural networks has undergone significant advancement in recent years, offering effective solutions across a wide spectrum of domains, ranging from engineering tasks to the analysis of social phenomena. Image-based inputs hold a particularly prominent role in these applications, with modern smartphones serving as cost-effective and widely accessible tools for capturing visual data. Road network monitoring and condition assessment represent domains that can considerably benefit from the integration of such technologies, as the collection and processing of relevant visual information can yield detailed and reliable insights into current conditions. Within the scope of the research, a pilot technological description is developed, outlining the process of field data acquisition using smartphones, followed by the processing of the resulting dataset through a deep neural network. The outcomes of this processing are presented using appropriate visualization methods, accompanied by an attempt to examine and validate their accuracy.