Information Theoretic Key Frame Selection for Action Recognition

Abstract

This paper presents an approach for human action recognition by finding the discriminative key frames from a video sequence and representing them with the distribution of local motion features and their spatiotemporal arrangements. In this approach, the key frames of the video sequence are selected by their discriminative power and represented by the local motion features detected in them and integrated from their temporal neighbors. In the key frame's representation, the spatial arrangements of the motion features are captured in a hierarchical spatial pyramid structure. By using frame by frame voting for the recognition, experiments have demonstrated improved performances over most of the other known methods on the popular benchmark data sets.

Recognising action as clouds of space-time interest points

“…Table 1 also compares our results with the existing approaches proposed recently, which are not restricted to interest points based methods. It shows that our results are close to the best result reported so far on each dataset, and outperforms most METHOD KTH WEZIMANN Our approach 93.17% 96.66% Fathi et al [5] 90.5% 100% Zhang et al [22] 91.33% 92.89% Kläser et al [9] 91.4% 84.3% Niebles et al [11] 83.3% 90.0% Dollar et al [3] 81.17% 85.2% Liu et al [10] 94.16% -Zhao et al [23] 91.17% -Gilbert et al [6] 89.92% -Savarese et al [17] 86.83% -Nowozin et al [12] 84.72% - recently proposed methods, especially those tested on both datasets.…”

Poselet Key-Framing: A Model for Human Activity Recognition

“…Keyframes: A number of approaches have proposed the use of keyframes as a representation. Carlson et al [5] use a single, manually selected, keyframe and shape matching to classify tennis strokes; [39] rank all frames based on holistic information theoretic measure to select the top 25% for classification using voting; [20] rely on spatio-temporal localization as pre-processing and use AdaBoost to select keyframes (making up from 13% to 20% of the sequence length). Unlike [5], we automatically select keyframes and return, not require as [20], spatial-temporal localization; our keyframe representation is also more compact utilizing fewer (up to 4, or 4% of the sequence) keyframes.…”

Learning Discriminative Key Poses for Action Recognition

“…To obtain visually distinct representations, Cooper and Foote [30] presented methods for key frame selection based on capturing the similarity to the represented segment and preserving the differences from other segments' key frames, so that different segments will have visually distinct representations. Zhao and Elgammal [31] developed an effective approach for action classification, in which they first described all the poses with the distribution of local motion features and their spatiotemporal arrangements and then selected a small set of most discriminative poses by comparing their discriminative power for each independent action. Zhuang et al [32] applied an unsupervised clustering method for key-pose selection.…”