The aim of our work herein was to design a photoplethysmographic (PPG) peak detection algorithm which automatically detect and discriminate various cardiac rhythms-normal sinus rhythms (NSR), premature atrial contraction (PAC), premature ventricle contraction (PVC), and atrial fibrillation (AF)-for PPG signals collected on smartwatch. Compared with peak detection algorithm designed for NSR, the novelty is that our proposed peak detection algorithm can accurately estimate heart rates (HR) among various arrhythmias, which enhances the accuracy of AF screening. Our peak detection method is composed of a sequential series of algorithms that are combined to discriminate various arrhythmias, as described above. Moreover, a novel Poincaré plot scheme is used to discriminate AF with Rapid Ventricular Response (RVR) from normal basal heart rate AF. Moreover, the method is also able to differentiate PAC/PVC from NSR and AF. Our results show that the proposed peak detection algorithm provides significantly lower average beat-to-beat estimation error (> 40% lower) and mean heart rate estimation error (> 50% lower) when compared to a traditional peak detection algorithm that is known to be accurate for NSR. Our new approach allows more accurate HR estimation as it can account for various arrhythmias which previous PPG peak detection algorithms were designed solely for NSR.