Abstract: Based on the gait data in the smart home environment, a method of data acquisition and emotion recognition is proposed, which is also privacy-friendly. The user’s gait dataset was collected through a capacitive sensored floor in a real home environment, and 18-dimensional gait features were extracted for analysis of emotion recognition. The GRU_FCN model was modified and optimized by the convolution layer and output layer to build the GRU_FCNPlus model. The results show that the accuracy of GRU_FCNPlus model in the four-category emotion recognition task was about 5% higher than that of the existing seven commonly used model algorithms. This method has a very broad application prospect in intelligent perception, human-room interaction, user experience and other aspects of smart home environment.