To reduce the impact of passengers' abnormal behavior on elevator safety and passenger personal safety, we propose a method for identifying dangerous behaviors of elevator passengers based on a temporal shift and time reinforcement module(TSTR module). Firstly, we propose a temporal shift and time reinforcement module to ensure the effective transmission of temporal-spatial features of human behavior information. Secondly, to verify the effectiveness of our proposed module, we replace the residual units in ResNet50 with the temporal shift and time reinforcement module and propose the TSTR-ResNet. Finally, considering that the data captured by elevator monitoring cameras contains a large amount of high-frequency noise, we use wavelet decomposition to extract low-frequency sub-band images as inputs to the neural network. Our method can accurately identify abnormal behaviors of elevator passengers, prevent elevator malfunctions caused by passenger behavior, and ensure passenger safety.