Motion Recognition by Higher Order Local Auto Correlation Features of Motion History Images

This paper proposes new features for motion recognition. Higher order local autocorrelation (HLAC) features are extracted from the motion history images (MHI). Since MHI calculated from the video images include important motion information, it is expected that HLAC features extracted from MHI have good properties for motion recognition. The proposed features were tested using image sequences of pitching in the baseball games. At first the pitchers were identified from the pitching motions by comparing the sequences of HLAC features using dynamic programming (DP) matching. The pitchers were recognized 100% correctly when the image size was 90times90 pixels. Then whether there was the runner on a base or not was identified. The recognition rate of the runners from the pitching motions was 96.7% when the image resolution was set to 25times25 pixels.