In this paper, we address the problem of controlling traffic lights at an intersection with a spatially distributed sensor network. We propose a sensor network architecture that does not depend on a centralized coordinator and we separate logically it into 4 levels of hierarchy. On this architecture, we define and evaluate through simulations an adaptive traffic light control algorithm. Based on two main objectives, this algorithm decides dynamically of the green lights sequences by selecting the movements composing each phase and its duration. Simulation results show that this algorithm, if properly tuned, has the capacity to reduce average waiting time at an intersection, while avoiding starvation.