A video event human-like concept learning method and a device, the method comprises: by using a random walk algorithm, performing a behavior division on a motion object in a video sample to obtain a behavior set (101), and performing a sub-behavior division on each behavior in the behavior set on the basis of a time pause point of the behavior to obtain a sub-behavior set (102); determining a time-space relation between each sub-behavior and other sub-behaviors in the sub-behavior set, and a priori probability of the time-space relation of each sub-behavior (103); using the time-space relation of each sub-behavior, the priori probability, and a preset factor set to compose a new video event and to learn so as to obtain the concept of a video event (104). Compared with the prior art, the above method enables to learn the concept of a video event, and express the concept through a simple and random plan, so as to realize a thinking which is similar to human beings for concept learning, thereby realizing the video event human-like concept learning.
展开▼