Event reconstruction for NOvA experiment is a critical step preceding further data analysis. We describe the complex NOvA reconstruction pipeline (containing several unsupervised learning techniques) with focus on the specific step of so-called "prong matching". In this step, we are combining 2D prongs (projections of particle trajectories) into 3D prong objects. In order to find the best matching 2D trajectory projections, we are using Kolmogorov-Smirnov homogeneity test on energy profiles. Moreover, we present the efficiency of the current matching process and propose some improvements on homogeneity testing.