The AI for Visual Vehicles Counting asset is a CNN-based algorithm able to estimate the traffic density and count the vehicles present in input images. We trained the network using an Unsupervised Domain Adaptation (UDA) strategy, where we suppose to have an annotated training set for a source domain, and we want to adapt the system to perform well in an unseen and unlabelled target domain. This class of algorithms is commonly referred to as Unsupervised Domain Adaptation.
Additional information: Monitoring vehicle flows in cities is a crucial issue to improve the urban environment and quality of life of citizens. Images are the best sensing modality to perceive and assess the flow of vehicles in large areas. Current technologies for vehicle counting in images hinge on large quantities of annotated data, preventing their scalability to city-scale as new cameras are added to the system. This is a recurrent problem when dealing with physical systems and a key research area in Machine Learning and AI. Most visual counting approaches rely on regression techniques to estimate a density map from the image, and where the final count is given by summing all pixel values. We propose a CNN-based algorithm for traffic density estimation and counting that can generalize to new data sources for which there are no annotations available. We achieve this generalization by exploiting an Unsupervised Domain Adaptation strategy, whereby a discriminator attached to the output forces similar density distribution in the target and source domains.