Automatic Imitation Assessment in Interaction

The MAHNOB Mimicry Database: A database of naturalistic human interactions

“…We consider face and head movements tracked by the state-of-the-arttrackers (i.e. the face tracker described in [25] and the head pose estimator described in [17]) and report on binary classification of video sequences into mimicry and non-mimicry categories based on the following widely-used methodology: two similarity-based methods (cross correlation as used in [22] and Generalised Time Warping [40]), and the state-of-the-art temporal classifier, Long Short Term Memory Recurrent Neural Network (LSTM-RNN) [32]. Performance of the methods is evaluated against the ground truth, representing human annotations of motor mimicry behaviour.…”

“…None of the works mentioned above attempt to compare their methods to ground truth of mimicry behaviour. Michelet et al [22] used a proprietary dataset for sequence classification into mimicry and non-mimicry classes. Their dataset contains 256 clips of posed, gross body movements, set against a uniform, static background.…”

“…ROC (receiver operating characteristic) curves were reported, giving a best performance of 0.920. Delaherche et al [7] proposed a similiar approach to that in [22], using the likelihood ratios between one-class SVMs as a distance measure between class distributions. They used the same posed data from [22].…”

“…Delaherche et al [7] proposed a similiar approach to that in [22], using the likelihood ratios between one-class SVMs as a distance measure between class distributions. They used the same posed data from [22]. They reported a maximum area-under-curve of 0.92, similarly to the results reported in [22].…”

Graph-based analysis of physical exercise actions

“…In the literature, we have encountered only few methods for automatic action quality assessment. These prior works have used Motion Capture (MoCap) data [6,10] or RGB video sequences [13]. A few approaches have been recently proposed based on skeleton tracking.…”