Accurate P wave detection is important for arrhythmia diagnosis and reliable P fiducial points delineation is necessary for electrocardiogram (ECG) analysis. Phasor transform is an effective method to assist automatic ECG fiducial points delineation, which maps each ECG sample into a phasor to enhance slight variations and preserves morphology and magnitude characteristics. Although phasor transform makes tiny waveforms easier to identify, more accurate ECG delineation methods still needs to be improved. This work improves the final results from four aspects: signal denoising, wave localization by phasor transformation, candidate fiducial points detection, and optimal selection. A simple notch filter together with intermittent median filters is applied for noise removal. The length of the search window and the degree of phasor transform are established through various trials. Especially, in addition to zero-crossing points of the phasor signal, the intersections of the phasor signal and the original ECG signal are taken into consideration as candidates, which make the most contribution to delineation results. Moreover, an optimal selection strategy for candidate fiducial points is applied throughout the whole process for the improvement of the delineation. For validation, the QT Database with 3194 p peak annotations from 105 records of two leads is adopted. As a result, we reached F1 = 94.67%, DER = 10.80% for p wave onset points detection, F1 = 93.56%, DER = 13.06% for p wave offset points detection, and F1 = 95.33%, DER = 9.46% for p wave peaks detection under a tolerance of 75 ms, which outperforms other reproducible works and their combinations.