Profiling video to visual track for preview

Video anatomy

“…In addition, dynamic scenes in a static field of view (FOV) may further reveal a variety of motions as directional motion (e.g., marathon crowds) or diversified motion (e.g., random walking in a shopping mall). In general, we can describe the camera kinematics [13], which yields typical camera works generating distinct optical flows that can be classified as diversified flow or directional flow. This categorization helps us design a general cutting strategy to obtain the profile of video clips.…”

“…For the simplicity, we assume here the major flow vector is horizontal, i.e., the camera motion is horizontal. From the simple camera motion such as zoom, rotation, and translation (abbreviated to Z, R, and T in bold) and their corresponding slice cutting [13], Figure 4 shows all the possible motion combinations. For example, a camera can zoom during translation (T+Z).…”

“…Figure 1 gives an example of such profiles from a linear camera motion. As an initial result, our profiling method has been working on simple camera actions such as zoom, pan, and translation [13]. This paper focuses on building a uniformed framework for the video volume cutting on general camera alignments and motions that are the combination of simple motions.…”

Localized temporal profile of surveillance video

“…For dynamic camera motion, a temporal profile has been generated from a moving sampling line to mainly summarize background [5,6,7]. However, the spatial information is lost when a temporal slice is cut from the spatial-temporal volume of video.…”