Currently, general human behavior datasets mainly focus on the visible light vision domain, such as the KTH dataset (including 25 persons, 6 actions), the WEIZMANN dataset (including 9 persons, 10 actions) and the INRIA XMAS multi-vision dataset (including 11 people, 14 actions), etc.
In terms of thermal infrared human action recognition dataset, Han and Bhanu's dataset includes 5 people and two types of actions (walking and running). The infrared action database (TADB) of Li Jianfu et al. contains 14 people and 12 kinds of daily life actions. The frame rate of this dataset is 25 frames per second, the frame size is 320x240, and each action is collected 4 times, with a total of 672 video sequences. An example of part of the behavior of this dataset is shown in the figure below.