Heart rate is an important indicator of human health. Detecting heart rate from face videos is a rapidly developing non-contact detection method, but the accuracy of the results is susceptible to noise such as changes of illumination and head movements of the subject. In order to eliminate the influence of noise and improve the accuracy of heart rate detection, a new idea is innovatively proposed: Illumination changes and head motion noise are used as targets for trend fitting and eliminated from the camera raw signal to achieve human heart rate signal extraction with high signal-to-noise ratio. In order to verify the feasibility of the proposed method, the value of the medical finger-clip pulse oximeter is used as the reference heart rate. By comparing the average signal-to-noise ratio at the reference heart rate and heart rate detection accuracy, the proposed method has increased by 9.60% and 37.19% compared with the POS method. Compared with the ICA algorithm, it has increased by 48.48% and 51.75%. Experimental results show that the new proposed method has better denoising ability and higher accuracy.