Some daily behaviors of the human body (such as walking, running, and jumping) usually exhibit approximately periodic characteristics within a certain period of time. Therefore, a cycle of behavior can be regarded as a natural unit of behavior. Therefore, by determining the cycle of different behaviors, the number of image frames involved in the behavior recognition process can be reduced, and the processing speed can be accelerated. Typical behavior cycle detection methods include image sequence correlation analysis, Fourier spectrum analysis and so on. Based on the behavior sequence images in a single cycle, various spatiotemporal templates can be constructed, and behavior features can be extracted from the templates to realize behavior recognition finally.

 

The spatiotemporal template can synthesize the behavioral images in one behavioral cycle into one image, which saves storage space and computational cost compared with behavioral recognition using multiple sequence images, and can record all changes of human contours in a behavioral cycle. When analyzing the behaviors with periodic characteristics, it is not sensitive to the time sequence of the behaviors, so it is natural to pay attention to the moving parts of the human body, that is, to highlight the moving characteristics. Therefore, spatiotemporal templates are an efficient compressed representation of human behavioral processes.