ABSTRACT
Video classification and categorization have always been challenging propositions as far as the computer vision and image processing communities are concerned. Tracking and localization of moving objects within video sequences, leading to proper classification of the sequences by means of detecting video shot boundaries, are prerequisites in several applications. Some of these applications include video surveillance, target tracking, defense, robotic maneuvering, and the entertainment and advertising industries, to name a few [7, 22]. Boreczky and Rowe [7] presented a comprehensive comparison of several techniques of video shot boundary detection. In a later reporting in [22], the authors provided yet another excellent as well as compact study on the pros and
for Video
cons of various methods of detection of change in video-shots in terms of their performances. Because video sequences entail a huge amount of data, estimation of the nature and degree of motion of the mobile objects within the motion scene is an uphill task in the image processing community. Moreover, video data is also continuous in nature. Hence, the commonly used estimation techniques in such a dynamic environment essentially comprise three approaches:
1. Using object centric feature matching in consecutive image frames [13], where Duncan and Chou exhibited the effectiveness of using optical flow for detection of video motion. Later on, the authors extended their work in [14].
