Cardiotocography is a measurement technique widely adopted to assess fetal well-being during both antepartum and intra-partum stages. It consists of two simultaneously acquired time-series: fetal heart rate and uterine activity. Since visual inspection analysis is gravely affected by intra- and inter-observer variability, recent literature is focusing on automated solutions. In this context, it is essential to provide a robust and accurate estimation of fetal heart rate to prevent from incorrect diagnosis. One of the major challenges is represented by the estimation of FHR baseline. The present paper proposes a novel algorithm capable of accurately estimating the baseline and correctly detecting possible pathological events, like heart rate accelerations or decelerations. In order to provide a thorough metrological characterization of the algorithm performances, we evaluate its accuracy by considering an experimental dataset.