at least one memory storing instructions; and

at least one processor configured to execute the instructions to perform operations comprising:

acquiring trajectory data representing trajectories of an object as time sequences of observed positions of the object;

assigning each trajectory to one of a plurality of groups for respectively corresponding behaviors of the object within a real-world space;

generating a generative model for each group; and

outputting the generative model generated for each group, wherein

the generative model for each group represents a common time sequence of transformations of velocity of the object from a previous time frame using a set of motion primitives common to all the groups and using a parameter related to a curvature of a left turn or a right turn of the object,

the transformations of the velocity of each group are invariant to a scene in which a camera has imaged the object and to a type of the camera,

the generative model further represents a time sequence of representative velocities by the common time sequence of the transformations of velocity of the object from the previous time frame, and

generating the generative model for each group comprises obtaining the time sequence of the representative velocities by performing marginalization on the trajectories assigned to the group, wherein

the set of motion primitives include going forward, going backward, going leftward, and going rightward, and

a relationship indicated by the following equations holds true between v_t and v_t-1, where v_t and v_t-1 are velocities in time frames adjacent to each other in order in time series and β is a parameter: