Phonocardiogram based auscultation is the most suitable cardiac examination technique for primary health care since heart sound can be captured and analyzed using a smart-phone and a digital stethoscope. The phonocardiogram signal provides, among others, valuable information about valve functioning of the heart. It is well known that many heart problems are associated with valve dysfunctions. Notably, the time differences between valves closure are very critical to diagnose some pathologies. Hence, the need of the correct detection of these instants. Up to now, this research problem represents a serious challenge. This Study takes place in this area of concern and targets to propose a greedy-based two-stage strategy to detect the instants of the heart valves closure. The first stage concerns the dictionary construction from the estimation of the impulse response functions associated to each heart valve. In the second stage, the instants of valves closures are identified by applying the Orthogonal Matching Pursuit algorithm alongside the constructed dictionaries. Simulations on both synthetic and real-life phonocardiogram signals are performed to validate the performance of the proposed two-stage approach in detecting the closure instants of the heart valves.