Activity recognition in unconstrained RGB-D video using 3D trajectories

Human action recognition has been a popular issue recently because it can be applied in many applications such as intelligent surveillance systems, human-robot interaction, and autonomous vehicle control. Human action recognition using RGB video is a challenging task because the learning of actions is easily affected by the cluttered background. To cope with this problem, the proposed method estimates 3D human poses first which can help remove the cluttered background and focus on the human body. In addition to the human poses, the proposed method also utilizes appearance features nearby the predicted joints to make our action prediction context-aware. Instead of using 3D convolutional neural networks as many action recognition approaches did, the proposed method uses a two-stream architecture that aggregates the results from skeleton-based and appearance-based approaches to do action recognition. Experimental results show that the proposed method achieved state-of-the-art performance on NTU RGB+D which is a large-scale dataset for human action recognition.

Regularized multi-task learning (MTL) algorithms have been exploited in the field of pattern recognition and computer vision gradually, which can fully excavate the relationships of different related tasks. Therefore, many dramatically favorable approaches based on regularized MTL have been proposed. In the past decades, although the promising results about human action recognition have been achieved, most of existing action recognition algorithms focus on action descriptors, single/multi-view and multi-modality action recognition, and few works are related with MTL, especial of lacking the systematic evaluation of existing MTL algorithms for human action recognition. Thus, in the paper, seven popular regularized MTL algorithms in which different actions are considered as different tasks, are systematically exploited on two public multi-view action datasets. In detail, dense trajectory features are firstly extracted for each view, and then the shared codebook are constructed for all views by k-means, and then each video is coded by the shared codebook. Moreover, according to different regularized MTL algorithms, all actions or part of actions are considered as related, and then these actions are set to different tasks in MTL. Further, the effectiveness of different number of training samples from different action views is also evaluated for MTL. Large scale experimental results show that: 1) Regularized MTL is very useful for action recognition which can dig the latent relationship among different actions; 2) Not of all human actions are related, if irrelative actions are put together in MTL, its performance will fall; 3) With the increase of the training samples from different views, the relationships about different actions can be fully exploited, and it promotes the accuracy improvement of action recognition.

We propose a unified method for recognizing human action and human related events in a realistic video. We use an efficient pipeline of (a) a 3D representation of the Improved Dense Trajectory Feature (DTF) and (b) Fisher Vector (FV). Further, a novel descriptor is proposed, capable of representing human actions and human related events based on the FV representation of the input video. The proposed unified descriptor is a 168-dimensional vector obtained from each video sequence by statistically analyzing the motion patterns of the 3D joint locations of the human body. The proposed descriptor is trained using binary Support Vector Machine (SVM) for recognizing human actions or human related events. We evaluate the proposed approach on two challenging action recognition datasets: UCF sports and CMU Mocap datasets. In addition to the two action recognition dataset, the proposed approach is tested on the Hollywood2 event recognition dataset. On all the benchmark datasets for both action and event recognition, the proposed approach has shown its efficacy compared to the state-of-the-art techniques.

Human activity recognition (AR) has begun to mature as a field, but for AR research to thrive, large, diverse, high quality, AR data sets must be publically available and AR methodology must be clearly documented and standardized. In the process of comparing our AR research to other efforts, however, we found that most AR data sets are sufficiently limited as to impact the reliability of existing research results, and that many AR research papers do not clearly document their experimental methodology and often make unrealistic assumptions. In this paper we outline problems and limitations with AR data sets and describe the methodology problems we noticed, in the hope that this will lead to the creation of improved and better documented data sets and improved AR experimental methodology. Although we cover a broad array of methodological issues, our primary focus is on an often overlooked factor, model type, which determines how AR training and test data are partitioned---and how AR models are evaluated. Our prior research indicates that personal, hybrid, and impersonal/universal models yield dramatically different performance [30], yet many research studies do not highlight or even identify this factor. We make concrete recommendations to address these issues and also describe our own publically available AR data sets.

Improved dense trajectory features have been successfully used in video-based action recognition problems, but their application to face processing is more challenging. In this paper, we propose a novel system that deals with the problem of emotion recognition in real-world videos, using improved dense trajectory, LGBP-TOP, and geometric features. In the proposed system, we detect the face and facial landmarks from each frame of a video using a combination of two recent approaches, and register faces by means of Procrustes analysis. The improved dense trajectory and geometric features are encoded using Fisher vectors and classification is achieved by extreme learning machines. We evaluate our method on the extended Cohn-Kanade (CK+) and EmotiW 2015 Challenge databases. We obtain state-of the-art results in both databases.

Human activity recognition is one of the most challenging research areas in human-machine interaction. Activity recognition by wearable sensors produces significant information about a person's functional ability and lifestyle. In this paper, we present an inertial sensors-based approach for activity and postural transition recognition. Efficient features are extracted from the sensor data. Then, the features are normalized using Z-score followed by Generalized Discriminant Analysis (GDA) to make them more robust. Finally, the features are trained with a Recurrent Neural Network (RNN) for activity and postural transition recognition. The proposed method recognizes 12 activities with an overall accuracy of 97%. In the activity recognition system, the multimodal input data is collected from only one accelerometer and one gyroscope sensor embedded in the user's smartphone. The proposed system can contribute in research fields such as human-robot interaction for practical applications such as eldercare in smart homes.

Local spatio-temporal interest points (STIPs) and the resulting features from RGB videos have been proven successful at activity recognition that can handle cluttered backgrounds and partial occlusions. In this paper, we propose its counterpart in depth video and show its efficacy on activity recognition. We present a filtering method to extract STIPs from depth videos (called DSTIP) that effectively suppress the noisy measurements. Further, we build a novel depth cuboid similarity feature (DCSF) to describe the local 3D depth cuboid around the DSTIPs with an adaptable supporting size. We test this feature on activity recognition application using the public MSRAction3D, MSRDailyActivity3D datasets and our own dataset. Experimental evaluation shows that the proposed approach outperforms state-of-the-art activity recognition algorithms on depth videos, and the framework is more widely applicable than existing approaches. We also give detailed comparisons with other features and analysis of choice of parameters as a guidance for applications.

Human Activity Recognition (HAR) based on accelerometer has become an important mobile application. Activity recognition however depends on X, Y, Z the Cartesian coordinate parameters. There are several approaches for activity recognition. Popular methods of activity recognition using accelerometer reading includes machine learning approach, rule based data mining approach, fuzzy inference approach etc. This paper compares activity recognition based on temporal pattern mining and ANFIS method for wearable sensor accelerometer and mobile accelerometer readings. Though the existing activity recognition using body worn accelerometer gives better accuracy it is found to be costly and consume more power. Hence this paper proposes an ANFIS based activity recognition with the available accelerometer in mobile phone.

Video-based human action recognition is one of the most important and challenging areas of research in the field of computer vision. Human action recognition has found many pragmatic applications in video surveillance, human-computer interaction, entertainment, autonomous driving, etc. Owing to the recent development of deep learning methods for human action recognition, the performance of action recognition has significantly enhanced for challenging datasets. Deep learning techniques are mainly used for recognizing actions in images and videos comprising of Euclidean data. A recent development in deep learning methods is the extension of these techniques to non-Euclidean data or graph data with many nodes and edges. Human body skeleton resembles a graph, therefore, the graph convolutional network (GCN) is applicable to the non-Euclidean body skeleton. In the past few years, GCN has emerged as an important tool for skeleton-based action recognition. Therefore, we conduct a survey using GCN methods for action recognition. Herein, we present a comprehensive overview of recent GCN techniques for action recognition, propose a taxonomy for the categorization of GCN techniques for action recognition, carry out a detailed study of the benchmark datasets, enlist relevant resources and open-source codes, and finally provide an outline for future research directions and trends. To the best of authors' knowledge, this is the first survey for action recognition using GCN techniques.

An automatic human action representation and recognition technique is proposed in this paper. Appearance-change problem due to human wearing dresses and body shapes is also investigated in this study for automatic human action recognition. A tuned eigenspace technique is proposed for automatic human posture and/or motion recognition that successfully overcome the preceding problems. We employ image pre-processing by Gaussian and Sobel edge filter, called the first stage tuning, for reducing a dress effect, and a mean eigenspace produced by taking a mean of the similar postures, called the second stage tuning, for avoiding the preceding problems. An eigenspace called a tuned eigenspace is obtained from the mentioned processes and it is used for further recognition of unfamiliar postures and actions. The proposed method is compared with a related technique and the robustness of this approach is presented.

Human action recognition from videos is a challenging task in computer vision. In recent years, histogram-based descriptors that are calculated along dense trajectories have shown promising results for human action recognition, but they usually ignore motion information of the tracking points, and the relationship between different motion variables is not well utilized. To address this issue, we propose a motion keypoint trajectory (MKT) approach and a trajectory-based covariance (TBC) descriptor, which is calculated along the motion keypoint trajectories. The proposed MKT approach tracks motion keypoints at multiple spatial scales and employs an optical flow rectification algorithm to reduce the influence of camera motions and thus achieves better performance than the improved dense trajectory (IDT) approach well known in the literature. In particular, MKT is faster than IDT, because MKT does not need to use human detection and extracts fewer trajectories than IDT. Furthermore, the TBC descriptor outperforms the classical histogram-based descriptors, such as the Histogram of Oriented Gradient, Histogram of Optical Flow and Motion Boundary Histogram. Experimental results on three challenging datasets (i.e., Olympic Sports, HMDB51 and UCF50) demonstrate that our approach is able to achieve better recognition performances than a number of state-of-the-art approaches.

Recent advancements in depth video sensors technologies have made human activity recognition (HAR) realizable for elderly monitoring applications. Although conventional HAR utilizes RGB video sensors, HAR could be greatly improved with depth video sensors which produce depth or distance information. In this paper, a depth-based life logging HAR system is designed to recognize the daily activities of elderly people and turn these environments into an intelligent living space. Initially, a depth imaging sensor is used to capture depth silhouettes. Based on these silhouettes, human skeletons with joint information are produced which are further used for activity recognition and generating their life logs. The life-logging system is divided into two processes. Firstly, the training system includes data collection using a depth camera, feature extraction and training for each activity via Hidden Markov Models. Secondly, after training, the recognition engine starts to recognize the learned activities and produces life logs. The system was evaluated using life logging features against principal component and independent component features and achieved satisfactory recognition rates against the conventional approaches. Experiments conducted on the smart indoor activity datasets and the MSRDailyActivity3D dataset show promising results. The proposed system is directly applicable to any elderly monitoring system, such as monitoring healthcare problems for elderly people, or examining the indoor activities of people at home, office or hospital.

Video and image analysis technologies have made human action recognition as an interesting field and used in many practical application systems such as visual surveillance systems, healthcare monitoring systems, 3D games and smart home. In this paper, we address the challenges of automatic tracking, detection and recognition of three-dimensional human pose-based actions from sequences of depth maps. Specifically, we have designed our proposed idea into a probabilistic framework: 1) dividing each action into meaningful ordered temporal segments, 2) using pixel-neighboring intensity difference approach to identify human shape region from the confused scenes, 3) introducing a robust spatiotemporal features to extract 3D joints information via local distance features and depth human shape information to get motion variation features and 4) finally these sets of features are merged together and their feature vectors are further reduced and discriminated to get optimal vectors for real-world applications. Next, the augmented features are enhanced and symbolized by Linde-Buzo-Gray clustering algorithm for better action recognition. With these symbols, each action hidden Markov model (HMM) is trained and tested for action recognition. The experimental results demonstrated our proposed approach on three challenging depth video datasets: IM-DepthActions, MSRAction3D and MSRDailyActivity3D, producing significantly improved results over the state-of-the-art methods: especially for those actions that are not easily discernible by the existing methods.

Human Activity Recognition (HAR) has become one of the most important research fields to achieve real-time monitoring of human activities for timely decision making in various applications like fall detection, elderly care etc. Now-a-days, most people use smartphones which come with various embedded inertial sensors like accelerometer and gyroscope to monitor acceleration and angular velocity. These smartphone-based sensors have proven to be cost-effective solution in identification of activities belonging to ADL (Activities of Daily Living). Various Machine Learning, Deep learning and hybrid models have been proposed and implemented for HAR. This paper also proposes various hybrid deep learning approaches which combine Deep Neural Networks with other models like LSTM (Long Short Term Memory) Model and GRU (Gated Recurrent Unit) for effective classification of engineered features from CNN (Convolutional Neural Network) Model. A novel architecture that integrates CNN with Random Forest Classifier (DeepCNN-RF) is proposed to add randomness to the model. The proposed models have been tested on publicly available HAR Datasets like UCI HAR and WISDM Activity Recognition Datasets. Experimental results show that the hybrid models outperform the state-of-the-art data mining, machine learning techniques in UCI HAR and WISDM with an overall maximum accuracy of 97.77% and 98.2% respectively.

In order to realize accurate recognition of human action, feature expression of motion information is a very crucial step. Aiming at the problem that the dense sampling used for action recognition will be affected by interference factors, such as camera motion and background information redundancy, this paper proposes the human action recognition method based on dense sampling of motion boundary and motion gradient histogram. Firstly, the dense sampling strategy based on motion boundary is incorporated into the improved dense sampling to eliminate a large number of invalid sampling points and reduce the number of trajectories. Next, in order to fully excavate the internal relationship of human movement between time and space, histograms of motion gradients based on time and space derivation is introduced to capture motion information in video, which is integrated with dense features to enhance the feature expression. The experiment results on two challenging datasets show that the proposed method improves the human action recognition accuracy effectively in the case of accelerating the speed of algorithm.