An accurate evaluation of the gravity acceleration from the timing of free fall of anti-hydrogen atoms in the GBAR experiment requires to account for obstacles surrounding the anti-matter source. These obstacles reduce the number of useful events but may improve accuracy since the edges of the shadow of obstacles on the detection chamber depends on gravity, bringing additional information on the value of $g$. We perform Monte Carlo simulations to obtain the dispersion and give a qualitative understanding of the results by analysing the statistics of events close to an edge. We also study the effect of specular quantum reflections of anti-hydrogen on surfaces and show that they do not degrade the accuracy that much.