Collaborative Learning for 3D Hand-Object Reconstruction and Compositional Action Recognition from Egocentric RGB Videos Using Superquadrics

The recognition of sports action is an important research subject, which is conducive to the improvement of athletes’ own level. To improve the accuracy of multi-modal data action recognition, based on the Transformer module, this study introduces a multi-head attention mechanism, fuses multi-modal data, and constructs a multi-stream structured object relationship inference network. Based on PointNet++ network and combining five different data fusion frameworks, a motion recognition model that integrates RGB data and 3D skeleton point cloud is constructed. The results showed that the Top-1 accuracy of multi-stream structured object relationship inference network was 42.5% and 42.7%, respectively, which was better than other algorithms. The accuracy of the multi-modal fusion model was improved by 15.6% and 5.1% compared with the single mode, and by 5.4% and 2.6% compared with the dual mode, which showed its superiority in the action recognition task. This showed that the fusion of multi-modal data can provide more abundant information, so as to improve the accuracy of action recognition. The accuracy of the action recognition model combining RGB data and 3D skeleton point cloud was 84.3%, 87.5%, 90.2%, 90.6% and 91.2% after the combination of different strategies, which effectively compensated for the problem of missing information in 3D skeleton point cloud and significantly improved the accuracy of action recognition. With a small amount of data, the Top-1 accuracy of the multi-stream structured object relationship inference network in this study was superior to other algorithms, showing its advantages in dealing with complex action recognition tasks. In addition, the action recognition model that fuses RGB data and 3D skeleton point cloud also achieved higher accuracy, which is better than other algorithms. This study can meet the needs of motion recognition in different scenarios and has certain reference value.

The task of action recognition in dark videos is useful in various scenarios, e.g., night surveillance and self-driving at night. Though progress has been made in action recognition task for videos in normal illumination, few have studied action recognition in the dark, partly due to the lack of sufficient datasets for such a task. In this paper, we explored the task of action recognition in dark videos. We bridge the gap of the lack of data by collecting a new dataset: the Action Recognition in the Dark (ARID) dataset. It consists of 3,784 video clips with 11 action categories. To the best of our knowledge, it is the first dataset focused on human actions in dark videos. To gain further understanding of our ARID dataset, we analyze our dataset in detail and showed its necessity over synthetic dark videos. Additionally, we benchmark the performance of current action recognition models on our dataset and explored potential methods for increasing their performances. We show that current action recognition models and frame enhancement methods may not be effective solutions for the task of action recognition in dark videos (data available at https://xuyu0010.github.io/arid).

Humans are able to intuitively deduce actions that took place between two states in observations via deductive reasoning. This is because the brain operates on a bidirectional communication model, which has radically improved the accuracy of recognition and prediction based on features connected to previous experiences. During the past decade, deep learning models for action recognition have significantly improved. However, deep neural networks struggle with these tasks on a smaller dataset for specific Action Recognition (AR) tasks. As with most action recognition tasks, the ambiguity of accurately describing activities in spatial-temporal data is a drawback that can be overcome by curating suitable datasets, including careful annotations and preprocessing of video data for analyzing various recognition tasks. In this study, we present a novel lightweight framework combining transfer learning techniques with a Conv2D LSTM layer to extract features from the pre-trained I3D model on the Kinetics dataset for a new AR task (Smart Baby Care) that requires a smaller dataset and less computational resources. Furthermore, we developed a benchmark dataset and an automated model that uses LSTM convolution with I3D (ConvLSTM-I3D) for recognizing and predicting baby activities in a smart baby room. Finally, we implemented video augmentation to improve model performance on the smart baby care task. Compared to other benchmark models, our experimental framework achieved better performance with less computational resources.

Human pose estimation and human activity recognition are two of the most researched domains in computer vision applications. Surveillance, human-computer interaction and video retrieval, all benefit from robust solutions of this domain. Due to disparities in inter-personal parameters, mobility efficiency, and recording configurations, the process is challenging. In this paper, deep learning algorithms are used for both pose estimation and activity recognition models. The pose estimation model generates an array of objects for each keypoint with their coordinated over a specific interval. This time series data is then consumed by the activity recognition model which was trained using a temporal convolutional neural network. The attained accuracy of the model was found to be 92.7% with a loss of 0.19.

Initializing the weights of a multilayer perceptron for activity and emotion recognition

Action recognition, a field within human-centered computing, is vital for identifying and understanding human actions, benefiting applications like surveillance, autonomous vehicles, and human-computer interaction. Artificial intelligence’s enduring goal is to develop robust models for perceiving and understanding the visual world around us. Deep neural networks have shown exceptional performance in various tasks, profoundly impacting real-world applications, including action prediction and recognition, which have advanced significantly in recent years. Most work in visual human action recognition focuses on single viewpoints or modalities from complete observation. Yet, the real significance lies in predicting future actions from incomplete observations to prevent real-world tragedies. With the availability of multiple cameras and data from multiple modalities (RGB, Depth, and Skeleton) available today, it becomes possible to model human action in multi-view and multi-modality context, minimizing the data loss due to occlusions and signal quality issues to improve recognition accuracy on the strength of state-of-the-art deep learning models. However, deep neural network models are susceptible to adversarial attacks, where imperceptible perturbations can compromise action recognition model performance. This thesis focuses on identifying latent vulnerabilities and proposing a defense mechanism against such threats in a multi-modality and multi-view setting. This work introduces an efficient and effective attack mechanism that perturbs skeleton data by targeting key joints and segments while employing a graph attention mechanism that learns the semantics to perturb other modalities. Additionally, an approach has been developed that not only adds noise but also alters the visual spatial structure of skeleton data through generative modeling. Furthermore, this dissertation introduces a defense mechanism known as the Collaborative Knowledge Distillation Network, which leverages graph attention and knowledge distillation techniques. This network leverages the knowledge from compromised multi-view data and integrates information from clean data to address incomplete observations and noisy action videos, enhancing the robustness of action recognition models for real-world applications.

Captions are text or logos superimposed on videos during a postproduction process. Caption detection in videos is useful for a variety of applications. For many applications, temporal consistency and stability is very important. Most of the prior work adopts certain post-processing procedures to smooth detected caption bounding boxes over time. Although these approaches mitigate the effect of the temporal inconsistency problem, they are unable to eliminate the problem. In this paper, we present a new caption detection algorithm that detects the 3D bounding boxes of caption regions in spatiotemporal volume space. 2D bounding boxes are then created by slicing the 3D bounding boxes. Since all the 2D bounding boxes corresponding to a caption area are sliced from one 3D bounding box, they are identical over time, thus ensuring temporal consistency of the result. The experiment results show that our new approach not only generates temporally consistent results but also results in higher detection accuracy.

Action recognition makes significant contributions to sports video analysis, especially for athletes’ training evaluations. For sports video analysis, the action information is mainly conveyed by human body parts’ temporal movement, and each of the parts has a unique importance to the action representation. Aiming to involve this point in action recognition, we propose a part-attention spatio-temporal graph convolutional network (PSGCN) to exploit the dynamic spatio-temporal information in a sports video; it learns the importance of different parts to emphasize the contribution on the task of action recognition. Specifically, PSGCN first divides the human body into six parts and extracts their convolutional neural network (CNN) features, as well as concatenating the global feature of the whole frame; it then utilizes a cross-part and cross-frame graph building module to formulate the graph correlation of the parts from different frames. Inspired by the larger temporal variation of the same part containing more action information, we further propose a part-attention (PA) learning module to estimate the importance of each part, which can strengthen the graph correlation to support a PA graph. Finally, PSGCN conducts a graph convolutional network on the learned PA spatio-temporal graph with the learned part CNN features, which can obtain the action representation for the given sports video. In addition, the whole network is optimized by two losses of PA and action classification. To perform the superiority of PSGCN, we carry out extensive experiments of our model compared with several state-of-the-art methods on widely used action recognition datasets, especially for sports action. The results reflect the advantages of the proposed PSGCN on sports video analysis.

Activity recognition aims to identify and predict human activities based on a series of sensor readings. In recent years, machine learning methods have become popular in solving activity recognition problems. A special difficulty for adopting machine learning methods is the workload to annotate a large number of sensor readings as training data. Labeling sensor readings for their corresponding activities is a time-consuming task. In practice, we often have a set of labeled training instances ready for an activity recognition task. If we can transfer such knowledge to a new activity recognition scenario that is different from, but related to, the source domain, it will ease our effort to perform manual labeling of training data for the new scenario. In this paper, we propose a transfer learning framework based on automatically learning a correspondence between different sets of sensors to solve this transfer-learning in activity recognition problem. We validate our framework on two different datasets and compare it against previous approaches of activity recognition, and demonstrate its effectiveness.

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.

Action recognition in videos is currently in the focus of scientific research due to improvements made in automatic analysis of static images and greater availability of processing power. The paper provides an overview of the key models and methods for action recognition that comprise human models and methods based on estimation of joint trajectories, silhouettes and template matching and spatio-temporal local descriptors. To deal with compound actions and activities, action semantic models are proposed with help of expert knowledge. Since the action recognition task is domain dependent, the methods and models are built and tested on domain specific databases. The paper provides an overview and description of recent video datasets that were created for developing action recognition methods, with an emphasis on datasets with additional modalities such as depth images or accelerometer data.

Human activity recognition is one of the most important research topics in the era of Internet of Things. To separate different activities given sensory data, we utilize a Minimal Distinguishing Subsequence (MDS) mining approach to efficiently find distinguishing patterns among different activities. We first transform the sensory data into a series of sensor triggering events and operate the MDS mining procedure afterwards. The gap constraints are also considered in the MDS mining. Given the multi-class nature of most activity recognition tasks, we modify the MDS mining approach from a binary case to a multi-class one to fit the need for multiple activity recognition. We also study how to select the best parameter set including the minimal and the maximal support thresholds in finding the MDSs for effective activity recognition. Overall, the prediction accuracy is 86.59% on the van Kasteren dataset which consists of four different activities for recognition.

Violence detection in videos is of great importance in many applications, ranging from teenagers protection to online media filtering and searching to surveillance systems. Typical methods mostly rely on hand-crafted features, which may lack enough discriminative capacity for the specific task of violent action recognition. Inspired by the good performance of deep models for human action recognition, we propose a novel method for detecting human violent behaviour in videos by integrating trajectory and deep convolutional neural networks, which takes advantage of hand-crafted features [21] and deep-learned features [23]. To evaluate this method, we carry out experiments on two different violence datasets: Hockey Fights dataset and Crowd Violence dataset. The results demonstrate the advantage of our method over state-of-the art methods on these datasets.

Deep appearance and motion learning for egocentric activity recognition

Human activity recognition is a very active research on pervasive computing and mobile health application. Many human activity systems based on inertial measurement unit (IMU) sensor data were proposed in the past few years. These systems mainly use IMU sensor placed on he torso and limbs to collect data and utilize supervised machine learning algorithms on sensor data. One main issue of these systems is that wearing multiple on-body IMU sensors may bring inconvenience to users' daily life. The other issue of these exiting methods is that an activity recognition model that is trained on a specific subject does not work well when being applied to predict another subject's activities since IMU activity data always carry information that is specific to the human subject who conducts the activities. In our work, inspired by the principle of domain adaption, we proposed a new deep-learning activity recognition model based on an adversarial network which can remove the subject-specific information within the IMU activity data and extract subject-independent features shared by the data collected on different subjects. We also for the first time use data collected from insole based IMU sensors on 8 participants for 5 common activities to build a new real world human activity dataset which can minimize the inconvenience for users to wear. We conducted experiments with our new real-world dataset. Results show that our subject independent activity recognition model outperforms state-of-art supervised learning techniques and eliminates the effects of individual differences between subjects successfully. The average recognition accuracy under the leave-one-out (L1O) condition achieves 99.0% which is higher than the performance of traditional human activity recognition system based on CNNs.