In this study, a remote photoplethysmography (rPPG) using a near-infrared camera image has been investigated to estimate the pulse rate in a dark environment. We propose a new face segmentation method using a skin boundary filter and a pixel-based shape mask filter to improve the rPPG performance. The proposed algorithms were evaluated by applying them to the skin of the whole face and the skin of the lower part of the face. The mean error rates of the estimated pulse rate in the lower part of the face and the whole face area were approximately 2.9% and 9.9%, respectively. Additionally, new dynamic noise cancelation techniques were proposed using a noise cancelation filter, the Kalman filter, and an adaptive filter to remove motion-related noise. Consequently, the mean error rate for 12 subjects drops to 2.0% when the proposed methods are combined. The rPPG signal showed robust results in motion artifact.