A tensor motion descriptor based on histograms of gradients and optical flow

Abstract

International audienceThis paper presents a new tensor motion descriptor only using optical flow and HOG3D information: no interest points are extracted and it is not based on a visual dictionary. We propose a new aggregation technique based on tensors. This is a double aggregation of tensor descriptors. The first one represents motion by using polynomial coefficients which approximates the optical flow. The other represents the accumulated data of all histograms of gradients of the video. The descriptor is evaluated by a classification of KTH, UCF11 and Hollywood2 datasets, using a SVM classifier. Our method reaches 93.2% of recognition rate with KTH, comparable to the best local ap- proaches. For the UCF11 and Hollywood2 datasets, our recognition achieves fairly competitive results compared to local and learning based approaches. Keywords: Global motion descriptor, optical flow, histogram of gradients, action recognitio

Efficient Imbalanced Multimedia Concept Retrieval by Deep Learning on Spark Clusters

“…Group Accuracy Perez et al (2012) 68. 9% Liu, Luo, & Shah (2009) 71.2% Mota et al (2014) 72.7% Our Framework 72.8%…”

“…In (Liu, Luo, & Shah, 2009), motion features based on the ROI (Region of Interest) estimation and AdaBoost were used to integrate all the heterogeneous yet complementary features for recognition. In (Mota et al, 2014), SVM was applied to a tensor motion descriptor with optical flow for action recognition. As shown in Table 6, our approach achieves the best accuracy rate among all the methods.…”

A Variable Size Block Matching Based Descriptor for Human Action Recognition

“…To represent the motion extracted, the most common structure used is the histogram [3], [9], [10], [11]. Histograms are interesting for video description as they are simple structures which carry a compact representation of the motion information.…”

“…Histograms are interesting for video description as they are simple structures which carry a compact representation of the motion information. In Mota et al [10], the final descriptor is an orientation tensor generated from a Histogram of Oriented Gradients (HOG). Tensors are robust mathematical tools and good aggregators.…”

“…If the motion captured in the histograms are too different from each other, we obtain an isotropic tensor which does not hold useful motion information. But, if accumulation results in an anisotropic tensor, it carries meaningful average motion information of the frame sequence [10].…”