Floods in dense urban areas propagate mainly through the streets, where the flow can be locally affected by elements of urban topography. This study aims at assessing the need of integrating detailed topography in numerical models when simulating urban floods. Acoustic Doppler Velocimetry and Large Scale Particle Image Velocimetry measurements in an experimental three branch junction representing a city crossroad are used to calibrate a numerical model solving the 2D shallow water equations. A constant eddy viscosity model proves to be accurate enough to calculate velocity fields, but such model requires a fine calibration against experimental data. Simulations run with this calibrated model are performed to study the impact of obstacles and sidewalks representative of urban areas. It is found that obstacles located in the downstream branch can highly perturb the velocities distribution downstream of the junction, whereas obstacles located in the upstream branches have less influence. The presence of sidewalks results in reduced flow section and higher velocities, but additional effects occur within and downstream of the junction. Simulations presented here show the need of considering detailed topography and elements of urban furniture if local velocities have to be represented.