Action Recognition in Videos Using Nonnegative Tensor Factorization

Abstract

Recognizing human actions is of vital interest in video surveillance or ambient assisted living. We consider an action as a sequence of body poses which are themselves a linear combination of body parts. In an offline procedure, nonnegative tensor factorization is used to extract basis images that represent body parts. The weighting coefficients are obtained by filtering a frame with the set of basis images. Since the basis images are obtained from nonnegative tensor factorization, they are separable and filtering can be implemented efficiently. The weighting coefficients encode dynamics and are used for action recognition. In the proposed action recognition framework, neither explicit detection and tracking of humans nor background subtraction are needed. Furthermore, for recognizing location specific actions, we implicitely take scene objects into account.

A tensor-based selection hyper-heuristic for cross-domain heuristic search

“…Having such a tensor filled with the data acquired from running the hyperheuristic for a short time and decomposing it, hopefully reveals the indices of low level heuristics which are performing well with the underlying hyper-heuristic and acceptance criteria. This is very similar to what has been done in [32], except that, instead of examining the video of human body motion and looking for different body parts moving in harmony, we examine the trace of a hyperheuristic (body motion) and look for low level heuristics (body parts) performing harmoniously. Naturally, our ultimate goal is to exploit this knowledge for improving the search process.…”

“…Similar claims have been registered in various research areas such as human action recognition in videos [32], hand written digit recognition [59], image compression [60], object recognition [58], gait recognition [54], Electroencephalogram (EEG) classification [37], Anomaly detection in streaming data [52], dimensionality reduction [39], tag recommendation systems [48] and Link…”

“…Those methods allow partitioning of the data into a set of more comprehensible sub-data which can be of specific use depending on the application according to various criteria. For example, in the field of computer vision, [30] and [32] separately show that factorization of a 3 rd -order tensor of a video sequence results in some interesting basic factors.…”

A tensor based hyper-heuristic for nurse rostering

“…It was shown in several studies (e.g. (Krausz and Bauckhage, 2010)) that the basic frame exhibits very interesting phenomena.…”

“…Since their introduction, tensorial approaches have been employed and greatly contributed to a variety of research areas such as computer vision (Vasilescu and Terzopoulos, 2002), video processing (Krausz and Bauckhage, 2010), data compression (Wang et al, 2009), Signal Processing (Cichocki et al, 2014) and web mining (Acar et al, 2009), (Zou et al, 2015). Various problems produce data of high order of dimensionality in nature and tensors, as multidimensional arrays, are fully suited to represent such data.…”