Classification of movies and television shows using motion

An innovative technique used for categorizing motion pictures and television shows based on motion activity is presented. First, scenes extracted from digitally stored movies and television programs are segmented into separate clips based on various editing effects such as hard cuts, fades and dissolves. Next, the initial frame of the sequence is automatically segmented into meaningful objects using color and texture features using any image segmentation algorithm chosen by the author. An object tracking algorithm utilizing motion is then applied to each Intra (I) frame and each Predicted (P) frame in the video sequence. Interesting objects undergoing the highest degree of motion are then automatically selected for additional analysis. The distance between corresponding object's centroids existing in adjacent frames is computed for each interesting object. These distances are then used in classifying the clip as being an action or non-action video.