Cyclic motion detection for motion based recognition

Ping-Sing Tsai, Mubarak Shah, Katharine Keiter, Takis Kasparis

Abstract: The motion of a walking person is analyzed by examining cycles in the movement. Cycles are detected using autocorrelation and Fourier transform techniques of the smoothed spatio-temporal curvature function of trajectories created by specific points on the object as it performs cyclic motion. A large impulse in the Fourier magnitude plot indicates the frequency at which cycles are occurring. Both synthetically generated and real walking sequences are analyzed for cyclic motion. The real sequences are then used in a motion based recognition application in which one complete cycle is stored as a model, and a matching process is performed using one cycle of an input trajectory.