Motion based retrieval of dynamic objects in videos

Abstract

Most existing video retrieval systems use low-level visual features such as color histogram, shape, texture, or motion. In this paper, we explore the use of higher-level motion representation for video retrieval of dynamic objects. We use three motion representations, which together can retrieve a large variety of motion patterns. Our approach works on top of a tracking unit and assumes that each dynamic object has been tracked and circumscribed in a minimal bounding box in each video frame. We represent the motion attributes of each object in terms of changes in the image context of its circumscribing box. The changes are described via motion templates [4], self-similarity plots [3], and image dynamics [9]. Initially, defined criteria of the retrieval process are interactively refined using relevance feedback from the user. Experimental results demonstrate the use of the proposed motion models in retrieving objects undergoing complex motion.

Action Recognition in Multimedia Streams

“…Using the motion information helps to segment relevant skin region with higher accuracy. Detecting periodic motion behaviour has become increasingly popular for retrieval in video [9,29,16]. The motion estimated here can be directly associated with periodicty of that skin region and thus a notion of illicit video [35].…”

Multimodal periodicity analysis for illicit content detection in videos

“…Periodic sounds in illicit materials are usually correlated to periodic motion occurring in the visual stream. Detecting periodic motion behaviour has become increasingly popular for retrieval in video [9,21,12]. Cutler [9] introduces similarity plot to illustrate the behaviour of natural periodic motion.…”

“…These similarity plots expose a confusion matrix of the cross-correlation of segmented dynamic objects over a particular duration. Liu et al [21] linearly combine similarity plots described above with the well known Motion History Image and Motion Energy Image along with an auto-regressive trajectory model to retrieve similar sequences from a repository of videos. Sport is the consideration of Fangxiang et al [12].…”

<title>Using video objects and relevance feedback in video retrieval</title>

“…Similar work, also operating on video rather than video keyframes, is reported in 6 where they automatically segment video frames into regions based on colour and texture, and then track the largest of these through a video sequence. Like the work of Liu et al in 5 they do not operate on segmented video objects but more like video blobs. In this approach a user query is not a segmented object but an object appearing in a query video clip.…”