The present invention belongs to the field of video images, and provides a geometric algebra-based unified model modeling method and a device for analyzing and processing videos. The method comprises the steps of obtaining the three-dimensional geometric algebra space of a video based on the space-domain information and the time-domain information of video images contained in the above video; constructing the motion vectors of pixel points in the three-dimensional geometric algebra space; in combination with the above motion vectors, constructing the apparent-motion vectors of the above pixel points, wherein the apparent-motion vectors of all the pixel points in the three-dimensional geometric algebra space represent an established video apparent and motion information unified model. According to the technical scheme of the invention, the geometric algebra, as a simple and efficient mathematical tool, is utilized for establishing a unified model between different dimensional geometries and the relations among the geometries based on the apparent information and the motion information of the video. Therefore, a detection algorithm for detecting space-time interest points is proposed based on the model. After that, the points of interest for clearly representing the change of the motion and containing the unique apparent information can be extracted in the time domain.