Abstract: The present disclosure relates to the field of emotion recognition technologies, and specifically, to a visual perception-based emotion recognition method. The method in the present disclosure includes the following steps: step 1: acquiring a face video by using a camera, performing face detection, and locating a region of interest (ROI); step 2: extracting color channel information of the ROI for factorization and dimension reduction; step 3: denoising and stabilizing signals and extracting physiological parameters including heart rate and breathing waveforms, and heart rate variability (HRV); step 4: extracting pre-training nonlinear features of the physiological parameters and performing modeling and regression of a twin network-based support vector machine (SVM) model; and step 5: reading a valence-arousal-dominance (VAD) model by using regressed three-dimensional information to obtain a specific emotional representation.