This work targets human action recognition in video. While recent methods typically represent actions by statistics of local video features, here we argue for the importance of a representation derived from human pose. To this end we propose a new Pose-based Convolutional Neural Network descriptor (P-CNN) for action recognition. The descriptor aggregates motion and appearance information along tracks of human body parts. We investigate different schemes of temporal aggregation and experiment with P-CNN features obtained both for automatically estimated and manually annotated human poses. We evaluate our method on the recent and challenging JHMDB and MPII Cooking datasets. For both datasets our method shows consistent improvement over the state of the art.
We propose an effective approach for spatio-temporal action localization in realistic videos. The approach first detects proposals at the frame-level and scores them with a combination of static and motion CNN features. It then tracks high-scoring proposals throughout the video using a tracking-by-detection approach. Our tracker relies simultaneously on instance-level and class-level detectors. The tracks are scored using a spatio-temporal motion histogram, a descriptor at the track level, in combination with the CNN features. Finally, we perform temporal localization of the action using a sliding-window approach at the track level. We present experimental results for spatio-temporal localization on the UCF-Sports, J-HMDB and UCF-101 action localization datasets, where our approach outperforms the state of the art with a margin of 15%, 7% and 12% respectively in mAP.
Unsupervised feature learning algorithms based on convolutional formulations of independent components analysis (ICA) have been demonstrated to yield state-of-the-art results in several action recognition benchmarks. However, existing approaches do not allow for the number of latent components (features) to be automatically inferred from the data in an unsupervised manner. This is a significant disadvantage of the state-of-the-art, as it results in considerable burden imposed on researchers and practitioners, who must resort to tedious cross-validation procedures to obtain the optimal number of latent features. To resolve these issues, in this paper we introduce a convolutional nonparametric Bayesian sparse ICA architecture for overcomplete feature learning from high-dimensional data. Our method utilizes an Indian buffet process prior to facilitate inference of the appropriate number of latent features under a hybrid variational inference algorithm, scalable to massive datasets. As we show, our model can be naturally used to obtain deep unsupervised hierarchical feature extractors, by greedily stacking successive model layers, similar to existing approaches. In addition, inference for this model is completely heuristics-free; thus, it obviates the need of tedious parameter tuning, which is a major challenge most deep learning approaches are faced with. We evaluate our method on several action recognition benchmarks, and exhibit its advantages over the state-of-the-art.
Human actions in video sequences are three-dimensional (3D) spatio-temporal signals characterizing both the visual appearance and motion dynamics of the involved humans and objects. Inspired by the success of convolutional neural networks (CNN) for image classification, recent attempts have been made to learn 3D CNNs for recognizing human actions in videos. However, partly due to the high complexity of training 3D convolution kernels and the need for large quantities of training videos, only limited success has been reported. This has triggered us to investigate in this paper a new deep architecture which can handle 3D signals more effectively. Specifically, we propose factorized spatio-temporal convolutional networks (FstCN) that factorize the original 3D convolution kernel learning as a sequential process of learning 2D spatial kernels in the lower layers (called spatial convolutional layers), followed by learning 1D temporal kernels in the upper layers (called temporal convolutional layers). We introduce a novel transformation and permutation operator to make factorization in FstCN possible. Moreover, to address the issue of sequence alignment, we propose an effective training and inference strategy based on sampling multiple video clips from a given action video sequence. We have tested FstCN on two commonly used benchmark datasets (UCF-101 and HMDB-51). Without using auxiliary training videos to boost the performance, FstCN outperforms existing CNN based methods and achieves comparable performance with a recent method that benefits from using auxiliary training videos.
Motion is a fundamental cue for scene analysis and human activity understan- ding in videos. It can be encoded in trajectories for tracking objects and for action recognition, or in form of flow to address behaviour analysis in crowded scenes. Each approach can only be applied on limited scenarios. We propose a motion-based system that represents the spatial and temporal features of the flow in terms of long-range trajectories. The novelty resides on the system formulation, its generic approach to handle scene variability and motion variations, motion integration from local and global representations, and the resulting long-range trajectories that overcome trajectory-based approach problems. We report the results and conclusions that state its pertinence on different scenarios, comparing and correlating the extracted trajectories of individual pedestrians, manually annotated. We also propose an evaluation framework and stress the diverse system characteristics that can be used for human activity tasks, namely on motion segmentation.
In this paper, a novel encoding scheme combining Fisher vector and bag-of-words encodings has been proposed for recognizing action in videos. The proposed Hyper-Fisher vector encoding is sum of local Fisher vectors which are computed based on the traditional Bag-of-Words (BoW) encoding. Thus, the proposed encoding is simple and yet an effective representation over the traditional Fisher Vector encoding. By extensive evaluation on challenging action recognition datasets, viz., Youtube, Olympic Sports, UCF50 and HMDB51, we show that the proposed Hyper-Fisher Vector encoding improves the recognition performance by around 2-3% compared to the improved Fisher Vector encoding. We also perform experiments to show that the performance of the Hyper-Fisher Vector is robust to the dictionary size of the BoW encoding.
Recent progress in using recurrent neural networks (RNNs) for image description has motivated the exploration of their application for video description. However, while images are static, working with videos requires modeling their dynamic temporal structure and then properly integrating that information into a natural language description. In this context, we propose an approach that successfully takes into account both the local and global temporal structure of videos to produce descriptions. First, our approach incorporates a spatial temporal 3-D convolutional neural network (3-D CNN) representation of the short temporal dynamics. The 3-D CNN representation is trained on video action recognition tasks, so as to produce a representation that is tuned to human motion and behavior. Second we propose a temporal attention mechanism that allows to go beyond local temporal modeling and learns to automatically select the most relevant temporal segments given the text-generating RNN. Our approach exceeds the current state-of-art for both BLEU and METEOR metrics on the Youtube2Text dataset. We also present results on a new, larger and more challenging dataset of paired video and natural language descriptions.
We address the problem of generating video features for action recognition. The spatial pyramid and its variants have been very popular feature models due to their success in balancing spatial location encoding and spatial invariance. Although it seems straightforward to extend spatial pyramid to the temporal domain (spatio-temporal pyramid), the large spatio-temporal diversity of unconstrained videos and the resulting significantly higher dimensional representations make it less appealing. This paper introduces the space-time extended descriptor, a simple but efficient alternative way to include the spatio-temporal location into the video features. Instead of only coding motion information and leaving the spatio-temporal location to be represented at the pooling stage, location information is used as part of the encoding step. This method is a much more effective and efficient location encoding method as compared to the fixed grid model because it avoids the danger of over committing to artificial boundaries and its dimension is relatively low. Experimental results on several benchmark datasets show that, despite its simplicity, this method achieves comparable or better results than spatio-temporal pyramid.
In this paper, a novel human action recognition technique from video is presented. Any action of human is a combination of several micro action sequences performed by one or more body parts of the human. The proposed approach uses spatio-temporal body parts movement (STBPM) features extracted from foreground silhouette of the human objects. The newly proposed STBPM feature estimates the movements of different body parts for any given time segment to classify actions. We also proposed a rule based logic named rule action classifier (RAC), which uses a series of condition action rules based on prior knowledge and hence does not required training to classify any action. Since we don't require training to classify actions, the proposed approach is view independent. The experimental results on publicly available Wizeman and MuHVAi datasets are compared with that of the related research work in terms of accuracy in the human action detection, and proposed technique outperforms the others.
Dictionary learning algorithms have been successfully used for both reconstructive and discriminative tasks, where an input signal is represented with a sparse linear combination of dictionary atoms. While these methods are mostly developed for single-modality scenarios, recent studies have demonstrated the advantages of feature-level fusion based on the joint sparse representation of the multimodal inputs. In this paper, we propose a multimodal task-driven dictionary learning algorithm under the joint sparsity constraint (prior) to enforce collaborations among multiple homogeneous/heterogeneous sources of information. In this task-driven formulation, the multimodal dictionaries are learned simultaneously with their corresponding classifiers. The resulting multimodal dictionaries can generate discriminative latent features (sparse codes) from the data that are optimized for a given task such as binary or multiclass classification. Moreover, we present an extension of the proposed formulation using a mixed joint and independent sparsity prior which facilitates more flexible fusion of the modalities at feature level. The efficacy of the proposed algorithms for multimodal classification is illustrated on four different applications -- multimodal face recognition, multi-view face recognition, multi-view action recognition, and multimodal biometric recognition. It is also shown that, compared to the counterpart reconstructive-based dictionary learning algorithms, the task-driven formulations are more computationally efficient in the sense that they can be equipped with more compact dictionaries and still achieve superior performance.
Deep ConvNets have shown its good performance in image classification tasks. However it still remains as a problem in deep video representation for action recognition. The problem comes from two aspects: on one hand, current video ConvNets are relatively shallow compared with image ConvNets, which limits its capability of capturing the complex video action information; on the other hand, temporal information of videos is not properly utilized to pool and encode the video sequences. Towards these issues, in this paper, we utilize two state-of-the-art ConvNets, i.e., the very deep spatial net (VGGNet) and the temporal net from Two-Stream ConvNets, for action representation. The convolutional layers and the proposed new layer, called frame-diff layer, are extracted and pooled with two temporal pooling strategy: Trajectory pooling and line pooling. The pooled local descriptors are then encoded with VLAD to form the video representations. In order to verify the effectiveness of the proposed framework, we conduct experiments on UCF101 and HMDB51 datasets. It achieves the accuracy of 93.78\% on UCF101 which is the state-of-the-art and the accuracy of 65.62\% on HMDB51 which is comparable to the state-of-the-art.
Human action recognition remains a challenging task due to the various sources of video data and large intra-class variations. It thus becomes one of the key issues in recent research to explore effective and robust representation to handle such challenges. In this paper, we propose a novel representation approach by constructing mid-level words in videos and encoding them on Riemannian manifold. Specifically, we first conduct a global alignment on the densely extracted low-level features to build a bank of corresponding feature groups, each of which can be statistically modeled as a mid-level word lying on some specific Riemannian manifold. Based on these mid-level words, we construct intrinsic Riemannian codebooks by employing K-Karcher-means clustering and Riemannian Gaussian Mixture Model, and consequently extend the Riemannian manifold version of three well studied encoding methods in Euclidean space, i.e. Bag of Visual Words (BoVW), Vector of Locally Aggregated Descriptors (VLAD), and Fisher Vector (FV), to obtain the final action video representations. Our method is evaluated in two tasks on four popular realistic datasets: action recognition on YouTube, UCF50, HMDB51 databases, and action similarity labeling on ASLAN database. In all cases, the reported results achieve very competitive performance with those most recent state-of-the-art works.
We make available to the community a new dataset to support action-recognition research. This dataset is different from prior datasets in several key ways. It is significantly larger. It contains streaming video with long segments containing multiple action occurrences that often overlap in space and/or time. All actions were filmed in the same collection of backgrounds so that background gives little clue as to action class. We had five humans replicate the annotation of temporal extent of action occurrences labeled with their class and measured a surprisingly low level of intercoder agreement. A baseline experiment shows that recent state-of-the-art methods perform poorly on this dataset. This suggests that this will be a challenging dataset to foster advances in action-recognition research. This manuscript serves to describe the novel content and characteristics of the LCA dataset, present the design decisions made when filming the dataset, and document the novel methods employed to annotate the dataset.
We present a review on the current state of publicly available datasets within the human action recognition community; highlighting the revival of pose based methods and recent progress of understanding person-person interaction modeling. We categorize datasets regarding several key properties for usage as a benchmark dataset; including the number of class labels, ground truths provided, and application domain they occupy. We also consider the level of abstraction of each dataset; grouping those that present actions, interactions and higher level semantic activities. The survey identifies key appearance and pose based datasets, noting a tendency for simplistic, emphasized, or scripted action classes that are often readily definable by a stable collection of sub-action gestures. There is a clear lack of datasets that provide closely related actions, those that are not implicitly identified via a series of poses and gestures, but rather a dynamic set of interactions. We therefore propose a novel dataset that represents complex conversational interactions between two individuals via 3D pose. 8 pairwise interactions describing 7 separate conversation based scenarios were collected using two Kinect depth sensors. The intention is to provide events that are constructed from numerous primitive actions, interactions and motions, over a period of time; providing a set of subtle action classes that are more representative of the real world, and a challenge to currently developed recognition methodologies. We believe this is among one of the first datasets devoted to conversational interaction classification using 3D pose features and the attributed papers show this task is indeed possible. The full dataset is made publicly available to the research community at www.csvision.swansea.ac.uk/converse.
Image and video classification research has made great progress through the development of handcrafted local features and learning based features. These two architectures were proposed roughly at the same time and have flourished at overlapping stages of history. However, they are typically viewed as distinct approaches. In this paper, we emphasize their structural similarities and show how such a unified view helps us in designing features that balance efficiency and effectiveness. As an example, we study the problem of designing efficient video feature learning algorithms for action recognition. We approach this problem by first showing that local handcrafted features and Convolutional Neural Networks (CNNs) share the same convolution-pooling network structure. We then propose a two-stream Convolutional ISA (ConvISA) that adopts the convolution-pooling structure of the state-of-the-art handcrafted video feature with greater modeling capacities and a cost-effective training algorithm. Through custom designed network structures for pixels and optical flow, our method also reflects distinctive characteristics of these two data sources. Our experimental results on standard action recognition benchmarks show that by focusing on the structure of CNNs, rather than end-to-end training methods, we are able to design an efficient and powerful video feature learning algorithm.
Nonlinear methods such as Deep Neural Networks (DNNs) are the gold standard for various challenging machine learning problems, e.g., image classification, natural language processing or human action recognition. Although these methods perform impressively well, they have a significant disadvantage, the lack of transparency, limiting the interpretability of the solution and thus the scope of application in practice. Especially DNNs act as black boxes due to their multilayer nonlinear structure. In this paper we introduce a novel methodology for interpreting generic multilayer neural networks by decomposing the network classification decision into contributions of its input elements. Although our focus is on image classification, the method is applicable to a broad set of input data, learning tasks and network architectures. Our method is based on deep Taylor decomposition and efficiently utilizes the structure of the network by backpropagating the explanations from the output to the input layer. We evaluate the proposed method empirically on the MNIST and ILSVRC data sets.
Most of human actions consist of complex temporal compositions of more simple actions. Action recognition tasks usually relies on complex handcrafted structures as features to represent the human action model. Convolutional Neural Nets (CNN) have shown to be a powerful tool that eliminate the need for designing handcrafted features. Usually, the output of the last layer in CNN (a layer before the classification layer -known as fc7) is used as a generic feature for images. In this paper, we show that fc7 features, per se, can not get a good performance for the task of action recognition, when the network is trained only on images. We present a feature structure on top of fc7 features, which can capture the temporal variation in a video. To represent the temporal components, which is needed to capture motion information, we introduced a hierarchical structure. The hierarchical model enables to capture sub-actions from a complex action. At the higher levels of the hierarchy, it represents a coarse capture of action sequence and lower levels represent fine action elements. Furthermore, we introduce a method for extracting key-frames using binary coding of each frame in a video, which helps to improve the performance of our hierarchical model. We experimented our method on several action datasets and show that our method achieves superior results compared to other state-of-the-arts methods.
Recurrent Neural Networks (RNNs) have had considerable success in classifying and predicting sequences. We demonstrate that RNNs can be effectively used in order to encode sequences and provide effective representations. The methodology we use is based on Fisher Vectors, where the RNNs are the generative probabilistic models and the partial derivatives are computed using backpropagation. State of the art results are obtained in two central but distant tasks, which both rely on sequences: video action recognition and image annotation. We also show a surprising transfer learning result from the task of image annotation to the task of video action recognition.
Recently, attempts have been made to collect millions of videos to train CNN models for action recognition in videos. However, curating such large-scale video datasets requires immense human labor, and training CNNs on millions of videos demands huge computational resources. In contrast, collecting action images from the Web is much easier and training on images requires much less computation. In addition, labeled web images tend to contain discriminative action poses, which highlight discriminative portions of a video's temporal progression. We explore the question of whether we can utilize web action images to train better CNN models for action recognition in videos. We collect 23.8K manually filtered images from the Web that depict the 101 actions in the UCF101 action video dataset. We show that by utilizing web action images along with videos in training, significant performance boosts of CNN models can be achieved. We then investigate the scalability of the process by leveraging crawled web images (unfiltered) for UCF101 and ActivityNet. We replace 16.2M video frames by 393K unfiltered images and get comparable performance.
Convolutional Architecture for Fast Feature Encoding (CAFFE) [11] is a software package for the training, classifying, and feature extraction of images. The UCF Sports Action dataset is a widely used machine learning dataset that has 200 videos taken in 720x480 resolution of 9 different sporting activities: diving, golf, swinging, kicking, lifting, horseback riding, running, skateboarding, swinging (various gymnastics), and walking. In this report we report on a caffe feature extraction pipeline of images taken from the videos of the UCF Sports Action dataset. A similar test was performed on overfeat, and results were inferior to caffe. This study is intended to explore the architecture and hyper parameters needed for effective static analysis of action in videos and classification over a variety of image datasets.