No abstract

We present a robust real-time vision-based system for vehicle tracking and categorization, developed for traffic flow surveillance. We propose a robust segmentation algorithm that detects foreground pixels corresponding to moving vehicles. Experimental results based on four large datasets show that our method can count and classify vehicles with a high level of performance (more than 98%).

Until humanity succeeds in massively producing clean energy to satisfy its inexhaustible needs, one of its biggest challenges is to save and use its resources as efficiently as possible. With outdoor lighting being responsible for 2% of worldwide electricity consumption, smart urban lighting has recently gained a lot of attention in this respect. As an integrated part of smart cities, smart urban lighting rests on the anal-ysis of sensed data to tackle highly dynamical problems. This sensed data shapes a representation of the environment in which the smart system will have to perform. To reduce problem complexity, distributed solu-tions commonly apply local lighting policies and therefore benefit from the knowledge of the geographical positioning of the relevant streetlights in the environment. In this paper, we propose an adaptive multiagent approach that aims at ensuring the robustness and coherence through time of the smart system's environment representation. Our approach leverages real time series data returned by streetlight sensors informing on vehicles and pedestrians traffic. We exploit this data to perform a structural reconstruction of the streetlight "fleet" topology without any a priori knowledge about its internal structure. We then ensure its cor-rectness through time by handling internal structure changes in order to continuously provide a coherent foundation for the smart lighting system to perform upon.