Grid-based Histogram of Oriented Optical Flow for analyzing movements on video data

Detection and recognition of object movements in a video is one of the research topics that are popular today. For the purposes of the analysis of the object movements in the video, the direction of movement is the important feature. In this study, we proposed a new method for determining the direction of movement using Histogram of Oriented Optical Flow (HOOF). We extract it locally at every N-by-N grid, not the entire frame. Direction movement is determined based on the value of HOOF on every grid. We classify the direction of movement in each grid into 12 directions. We use a video from UMN datasets for testing the proposed method. The experiment results show the value of False Positive Per Grid (FPPG) is 28.32%, and False Negative Per Grid (FNPG) is 4.08%. It proved that the use of Grid-based HOOF for analyzing movements on video data is good enough and can be improved in the future studies.