Probabilistic subspace-based learning of shape dynamics modes for multi-view action recognition

Abstract

We propose a human action recognition algorithm by capturing a compact signature of shape dynamics from multi-view videos. First, we compute R transforms and its temporal velocity on action silhouettes from multiple views to generate a robust low level representation of shape. The spatio-temporal shape dynamics across all the views is then captured by fusion of eigen and multiset partial least squares modes. This provides us a lightweight signature which is classified using a probabilistic subspace similarity technique by learning inter-action and intra-action models. Quantitative and qualitative results of our algorithm are reported on MuHAVi a publicly available multi-camera multiaction dataset.

Human action recognition with sparse classification and multiple‐view learning

“…A joint bag‐of‐words histogram might be built with local feature descriptors extracted from each one of the camera views (Wu et al ., ), but a better performance is reported when other fusion strategies are employed. Projections maximizing cross‐covariance have been learned to combine ‐transform derivatives extracted from each camera view (Karthikeyan et al ., ). Two‐level linear discriminant analysis has been employed to learn silhouette projections maximizing action class separability (Iosifidis et al ., ).…”

Modeling and recognizing action contexts in persons using sparse representation

“…In [43], Delaitre et al proposed a statistical model which integrates order-less person-object interaction responses to recognize actions in still images by calculating spatial co-occurrences of individual body parts and objects. To achieve the property of view invariance, Karthikeyan et al [8] computed the R transforms on action silhouettes from multiple views and then proposed a probabilistic subspace similarity technique to recognize actions by learning their inter-action and intra-action models. However, when recognizing an action, the scheme requires that all action sequences captured at different views should be collected together simultaneously.…”

“…There have been many approaches [8,9,[13][14][15][16][17][18]42,44] proposed in the literature for video-based human movement analysis. For example, in [52], Fengjun and Nevatia used a posture-based scheme to convert frames to strings from which actions were then classified into different types using the action nets.…”

Human movement analysis around a view circle using time-order similarity distributions

“…Furthermore, Cuzzolin [23] built different bilinear HMMs to analyze human gaits from multiple views. Karthikeyan, Gaur, and Manjunath [24] computed R transforms on action silhouettes and then proposed a probabilistic subspace similarity technique to recognize actions from multiple views by learning their inter-action and intra-action models. However, this scheme requires all action sequences from different views must be collected together before recognize an action.…”