Demand estimation is an important step for multiple applications in urban studies. However, the level of accuracy required depends on the objective of the study. In dynamic traffic microsimulation, the estimation of demand needs to be accurate, as it is intended to describe individuals' trips on a very small-scale. In this case, poor estimation of trip initiation, path, and endings could result in the wrong estimation of the city-block scale's traffic state. Estimating demand at a large-scale with high-resolution is not only very challenging because it requires a large volume of data from multiple sources, but the underlying mathematical problem is considerable and thus hard to solve. In this paper, we address the issue of trip starts and ends when modeling large perimeters. We propose to enhance the location of trip initiation and termination by merging heterogeneous and large public datasets. To do so, we develop a series of algorithms that identify fine-mesh areas where trips could reliably start or end and we share the estimated demand within these sub-areas, following the distribution of trip purposes (Home, Work, Shop, etc.). The method is deployed in Lyon city, France, and validated on an extraction of it. Micro-simulation results show that the demand, once accurately distributed, changes the overall network's performance, confirming the significant influence of trip endings and starts on the overall traffic dynamics.