Action Recognition Framework Research Articles

Hypergraph-Based Multi-View Action Recognition Using Event Cameras.

Action recognition from video data forms a cornerstone with wide-ranging applications. Single-view action recognition faces limitations due to its reliance on a single viewpoint. In contrast, multi-view approaches capture complementary information from various viewpoints for improved accuracy. Recently, event cameras have emerged as innovative bio-inspired sensors, leading to advancements in event-based action recognition. However, existing works predominantly focus on single-view scenarios, leaving a gap in multi-view event data exploitation, particularly in challenges like information deficit and semantic misalignment. To bridge this gap, we introduce HyperMV, a multi-view event-based action recognition framework. HyperMV converts discrete event data into frame-like representations and extracts view-related features using a shared convolutional network. By treating segments as vertices and constructing hyperedges using rule-based and KNN-based strategies, a multi-view hypergraph neural network that captures relationships across viewpoint and temporal features is established. The vertex attention hypergraph propagation is also introduced for enhanced feature fusion. To prompt research in this area, we present the largest multi-view event-based action dataset THUMV-EACT-50, comprising 50 actions from 6 viewpoints, which surpasses existing datasets by over tenfold. Experimental results show that HyperMV significantly outperforms baselines in both cross-subject and cross-view scenarios, and also exceeds the state-of-the-arts in frame-based multi-view action recognition.

A Novel Active Learning Framework for Cross-Subject Human Activity Recognition from Surface Electromyography.

Wearable sensor-based human activity recognition (HAR) methods hold considerable promise for upper-level control in exoskeleton systems. However, such methods tend to overlook the critical role of data quality and still encounter challenges in cross-subject adaptation. To address this, we propose an active learning framework that integrates the relation network architecture with data sampling techniques. Initially, target data are used to fine tune two auxiliary classifiers of the pre-trained model, thereby establishing subject-specific classification boundaries. Subsequently, we assess the significance of the target data based on classifier discrepancy and partition the data into sample and template sets. Finally, the sampled data and a category clustering algorithm are employed to tune model parameters and optimize template data distribution, respectively. This approach facilitates the adaptation of the model to the target subject, enhancing both accuracy and generalizability. To evaluate the effectiveness of the proposed adaptation framework, we conducted evaluation experiments on a public dataset and a self-constructed electromyography (EMG) dataset. Experimental results demonstrate that our method outperforms the compared methods across all three statistical metrics. Furthermore, ablation experiments highlight the necessity of data screening. Our work underscores the practical feasibility of implementing user-independent HAR methods in exoskeleton control systems.

A novel multi-stream hand-object interaction network for assembly action recognition

PurposeAssembly action recognition plays an important role in assembly process monitoring and human-robot collaborative assembly. Previous works overlook the interaction relationship between hands and operated objects and lack the modeling of subtle hand motions, which leads to a decline in accuracy for fine-grained action recognition. This paper aims to model the hand-object interactions and hand movements to realize high-accuracy assembly action recognition.Design/methodology/approachIn this paper, a novel multi-stream hand-object interaction network (MHOINet) is proposed for assembly action recognition. To learn the hand-object interaction relationship in assembly sequence, an interaction modeling network (IMN) comprising both geometric and visual modeling is exploited in the interaction stream. The former captures the spatial location relation of hand and interacted parts/tools according to their detected bounding boxes, and the latter focuses on mining the visual context of hand and object at pixel level through a position attention model. To model the hand movements, a temporal enhancement module (TEM) with multiple convolution kernels is developed in the hand stream, which captures the temporal dependences of hand sequences in short and long ranges. Finally, assembly action prediction is accomplished by merging the outputs of different streams through a weighted score-level fusion. A robotic arm component assembly dataset is created to evaluate the effectiveness of the proposed method.FindingsThe method can achieve the recognition accuracy of 97.31% and 95.32% for coarse and fine assembly actions, which outperforms other comparative methods. Experiments on human-robot collaboration prove that our method can be applied to industrial production.Originality/valueThe author proposes a novel framework for assembly action recognition, which simultaneously leverages the features of hands, objects and hand-object interactions. The TEM enhances the representation of dynamics of hands and facilitates the recognition of assembly actions with various time spans. The IMN learns the semantic information from hand-object interactions, which is significant for distinguishing fine assembly actions.

Data augmentation aided excavator activity recognition using deep convolutional conditional generative adversarial networks

Excavator activity recognition is a crucial task with significant practical applications such as assessing production efficiency, injury prevention, and enabling intelligent control. However, implementing deep learning-based activity recognition methods presents considerable challenges due to the requirement of large training datasets. And sensor-based activity recognition demands more convenient and reliable sensors. This paper proposes a generalized data augmentation framework for excavator activity recognition. Main pump pressure and flow signals are generated using Deep Convolutional Conditional Generative Adversarial Networks (DC-CGAN). The proposed framework integrates Long Short-Term Memory (LSTM) and Convolutional Neural Network (CNN) architectures for activity recognition. Results indicate that the framework outperforms traditional data augmentation methods regarding the quality of generated data and shallow and single deep networks regarding model accuracy and generalization. Additionally, the framework was implemented on three sizes of excavators and a wheel loader, demonstrating excellent versatility.

Hybrid attentive prototypical network for few-shot action recognition

Most previous few-shot action recognition works tend to process video temporal and spatial features separately, resulting in insufficient extraction of comprehensive features. In this paper, a novel hybrid attentive prototypical network (HAPN) framework for few-shot action recognition is proposed. Distinguished by its joint processing of temporal and spatial information, the HAPN framework strategically manipulates these dimensions from feature extraction to the attention module, consequently enhancing its ability to perform action recognition tasks. Our framework utilizes the R(2+1)D backbone network, coupling the extraction of integrated temporal and spatial features to ensure a comprehensive understanding of video content. Additionally, our framework introduces the novel Residual Tri-dimensional Attention (ResTriDA) mechanism, specifically designed to augment feature information across the temporal, spatial, and channel dimensions. ResTriDA dynamically enhances crucial aspects of video features by amplifying significant channel-wise features for action distinction, accentuating spatial details vital for capturing the essence of actions within frames, and emphasizing temporal dynamics to capture movement over time. We further propose a prototypical attentive matching module (PAM) built on the concept of metric learning to resolve the overfitting issue common in few-shot tasks. We evaluate our HAPN framework on three classical few-shot action recognition datasets: Kinetics-100, UCF101, and HMDB51. The results indicate that our framework significantly outperformed state-of-the-art methods. Notably, the 1-shot task, demonstrated an increase of 9.8% in accuracy on UCF101 and improvements of 3.9% on HMDB51 and 12.4% on Kinetics-100. These gains confirm the robustness and effectiveness of our approach in leveraging limited data for precise action recognition.

FTAN: Frame-to-frame temporal alignment network with contrastive learning for few-shot action recognition

Most current few-shot action recognition approaches follow the metric learning paradigm, measuring the distance of any sub-sequences (frames, any frame combinations or clips) between different actions for classification. However, this disordered distance metric between action sub-sequences ignores the long-term temporal relations of actions, which may result in significant metric deviations. What's more, the distance metric suffers from the distinctive temporal distribution of different actions, including intra-class temporal offsets and inter-class local similarity. In this paper, a novel few-shot action recognition framework, Frame-to-frame Temporal Alignment Network (FTAN), is proposed to address the above challenges. Specifically, an attention-based temporal alignment (ATA) module is devised to calculate the distance between corresponding frames of different actions along the temporal dimension to achieve frame-to-frame temporal alignment. Meanwhile, the Temporal Context module (TCM) is proposed to increase inter-class diversity by enriching the frame-level feature representation, and the Frames Cyclic Shift Module (FCSM) performs frame-level temporal cyclic shift to reduce intra-class inconsistency. In addition, we present temporal and global contrastive objectives to assist in learning discriminative and class-agnostic visual features. Experimental results show that the proposed architecture achieves state-of-the-art on HMDB51, UCF101, Something-Something V2 and Kinetics-100 datasets.

An end-to-end hand action recognition framework based on cross-time mechanomyography signals

The susceptibility of mechanomyography (MMG) signals acquisition to sensor donning and doffing, and the apparent time-varying characteristics of biomedical signals collected over different periods, inevitably lead to a reduction in model recognition accuracy. To investigate the adverse effects on the recognition results of hand actions, a 12-day cross-time MMG data collection experiment with eight subjects was conducted by an armband, then a novel MMG-based hand action recognition framework with densely connected convolutional networks (DenseNet) was proposed. In this study, data from 10 days were selected as a training subset, and the remaining data from another 2 days were used as a test set to evaluate the model’s performance. As the number of days in the training set increases, the recognition accuracy increases and becomes more stable, peaking when the training set includes 10 days and achieving an average recognition rate of 99.57% (± 0.37%). In addition, part of the training subset is extracted and recombined into a new dataset and the better classification performances of models can be achieved from the test set. The method proposed effectively mitigates the adverse effects of sensor donning and doffing on recognition results.

AGAR - Attention Graph-RNN for Adaptative Motion Prediction of Point Clouds of Deformable Objects

This article focuses on motion prediction for point cloud sequences in the challenging case of deformable 3D objects, such as human body motion. First, we investigate the challenges caused by deformable shapes and complex motions present in this type of representation, with the ultimate goal of understanding the technical limitations of state-of-the-art models. From this understanding, we propose an improved architecture for point cloud prediction of deformable 3D objects. Specifically, to handle deformable shapes, we propose a graph-based approach that learns and exploits the spatial structure of point clouds to extract more representative features. Then, we propose a module able to combine the learned features in a adaptative manner according to the point cloud movements. The proposed adaptative module controls the composition of local and global motions for each point, enabling the network to model complex motions in deformable 3D objects more effectively. We tested the proposed method on the following datasets: MNIST moving digits, the Mixamo human bodies motions [ 15 ], JPEG [ 5 ] and CWIPC-SXR [ 32 ] real-world dynamic bodies. Simulation results demonstrate that our method outperforms the current baseline methods given its improved ability to model complex movements as well as preserve point cloud shape. Furthermore, we demonstrate the generalizability of the proposed framework for dynamic feature learning by testing the framework for action recognition on the MSRAction3D dataset [ 19 ] and achieving results on par with state-of-the-art methods.

Human action recognition with transformer based on convolutional features

As one of the key research directions in the field of computer vision, human action recognition has a wide range of practical application values and prospects. In the fields of video surveillance, human-computer interaction, sports analysis, and healthcare, human action recognition technology shows a broad application prospect and potential. However, the diversity and complexity of human actions bring many challenges, such as handling complex actions, distinguishing similar actions, coping with changes in viewing angle, and overcoming occlusion problems. To address the challenges, this paper proposes an innovative framework for human action recognition. The framework combines the latest pose estimation algorithms, pre-trained CNN models, and a Vision Transformer to build an efficient system. The first step involves utilizing the latest pose estimation algorithm to accurately extract human pose information from real RGB image frames. Then, a pre-trained CNN model is used to perform feature extraction on the extracted pose information. Finally, the Vision Transformer model is applied for fusion and classification operations on the extracted features. Experimental validation is conducted on two benchmark datasets, UCF 50 and UCF 101, to demonstrate the effectiveness and efficiency of the proposed framework. The applicability and limitations of the framework in different scenarios are further explored through quantitative and qualitative experiments, providing valuable insights and inspiration for future research.

Temporal cues enhanced multimodal learning for action recognition in RGB-D videos

Action recognition is an important and active research direction in computer vision, where temporal modeling is critical for action representation. Generally, unimodal methods that use only RGB or skeleton modality for human action recognition have their limitations, e.g., information redundancy/environment noise of RGB video modality, and spatial interaction deficiency of skeleton modality. In this paper, we present a novel multimodal learning approach based on RGB and skeleton modalities for action recognition in RGB-D videos. Specifically, we (1) transfer skeleton knowledge to RGB video for effective video compression, which produces the informative action image from raw RGB video, (2) introduce the temporal cues enhancement module to adequately learn the spatiotemporal representation for action classification, and (3) propose a multi-level multimodal co-learning framework for human action recognition in RGB-D videos. Experimental results on NTU RGB+D, PKU-MMD, and N-UCLA datasets demonstrate the effectiveness of the proposed multimodal learning method.

Multi-head CNN-based activity recognition and its application on chest-mounted sensor-belt

In recent years, a great deal of research has been done on the identification, monitoring, and classification of human activities. Human activity recognition (HAR) is a term commonly used to describe the automatic identification of physical activities. For activity recognition, there are primarily vision-based and sensor-based methods available. The computer vision-based method is generally effective in lab settings, but because of clutter, fluctuating light levels, and contrast, it may not perform well in real-world scenarios. Continuous monitoring and analysis of physiological signals obtained from heterogeneous sensors attached to an individual’s body is required to realise sensor-based HAR systems. Most of the previous research in human activity recognition (HAR) is biased along with feature engineering and pre-processing which requires a good amount of domain knowledge. Application-specific modelling and time-taking methods are involved in these approaches. In this work, the multi-head convolutional neural network-based human activity recognition framework is proposed where automatic feature extraction and classification are involved in the form of an end-to-end classification approach. Experiments of this approach are performed by taking raw wearable sensor data with few pre-processing steps and without the involvement of a handcrafted feature extraction technique. 99.23% and 93.55% accuracy are obtained on the WISDM and UCI-HAR datasets which denoted the much improvement in the assessment of HAR over other similar approaches. The model is also tested on locally collected data from a chest mounted belt with fabric sensors and an accuracy of 87.14% has been achieved on that data.

AP-TransNet: a polarized transformer based aerial human action recognition framework

AP-TransNet: a polarized transformer based aerial human action recognition framework

A Multimodal, Multi-Task Adapting Framework for Video Action Recognition

Recently, the rise of large-scale vision-language pretrained models like CLIP, coupled with the technology of Parameter-Efficient FineTuning (PEFT), has captured substantial attraction in video action recognition. Nevertheless, prevailing approaches tend to prioritize strong supervised performance at the expense of compromising the models' generalization capabilities during transfer. In this paper, we introduce a novel Multimodal, Multi-task CLIP adapting framework named M2-CLIP to address these challenges, preserving both high supervised performance and robust transferability. Firstly, to enhance the individual modality architectures, we introduce multimodal adapters to both the visual and text branches. Specifically, we design a novel visual TED-Adapter, that performs global Temporal Enhancement and local temporal Difference modeling to improve the temporal representation capabilities of the visual encoder. Moreover, we adopt text encoder adapters to strengthen the learning of semantic label information. Secondly, we design a multi-task decoder with a rich set of supervisory signals, including the original contrastive learning head, a cross-modal classification head, a cross-modal masked language modeling head, and a visual classification head. This multi-task decoder adeptly satisfies the need for strong supervised performance within a multimodal framework. Experimental results validate the efficacy of our approach, demonstrating exceptional performance in supervised learning while maintaining strong generalization in zero-shot scenarios.

Shots segmentation-based optimized dual-stream framework for robust human activity recognition in surveillance video

Nowadays, for controlling crime, surveillance cameras are typically installed in all public places to ensure urban safety and security. However, automating Human Activity Recognition (HAR) using computer vision techniques faces several challenges such as lowlighting, complex spatiotemporal features, clutter backgrounds, and inefficient utilization of surveillance system resources. Existing attempts in HAR designed straightforward networks by analyzing either spatial or motion patterns resulting in limited performance while the dual streams methods are entirely based on Convolutional Neural Networks (CNN) that are inadequate to learning the long-range temporal information for HAR. To overcome the above-mentioned challenges, this paper proposes an optimized dual stream framework for HAR which mainly consists of three steps. First, a shots segmentation module is introduced in the proposed framework to efficiently utilize the surveillance system resources by enhancing the lowlight video stream and then it detects salient video frames that consist of human. This module is trained on our own challenging Lowlight Human Surveillance Dataset (LHSD) which consists of both normal and different levels of lowlighting data to recognize humans in complex uncertain environments. Next, to learn HAR from both contextual and motion information, a dual stream approach is used in the feature extraction. In the first stream, it freezes the learned weights of the backbone Vision Transformer (ViT) B-16 model to select the discriminative contextual information. In the second stream, ViT features are then fused with the intermediate encoder layers of FlowNet2 model for optical flow to extract a robust motion feature vector. Finally, a two stream Parallel Bidirectional Long Short-Term Memory (PBiLSTM) is proposed for sequence learning to capture the global semantics of activities, followed by Dual Stream Multi-Head Attention (DSMHA) with a late fusion strategy to optimize the huge features vector for accurate HAR. To assess the strength of the proposed framework, extensive empirical results are conducted on real-world surveillance scenarios and various benchmark HAR datasets that achieve 78.6285%, 96.0151%, and 98.875% accuracies on HMDB51, UCF101, and YouTube Action, respectively. Our results show that the proposed strategy outperforms State-of-the-Art (SOTA) methods. The proposed framework gives superior performance in HAR, providing accurate and reliable recognition of human activities in surveillance systems.

Residual deep gated recurrent unit-based attention framework for human activity recognition by exploiting dilated features

Residual deep gated recurrent unit-based attention framework for human activity recognition by exploiting dilated features

A human activity recognition framework in videos using segmented human subject focus

A human activity recognition framework in videos using segmented human subject focus

Subsampled Randomized Hadamard Transformation-based Ensemble Extreme Learning Machine for Human Activity Recognition

Extreme Learning Machine (ELM) is becoming a popular learning algorithm due to its diverse applications, including Human Activity Recognition (HAR). In ELM, the hidden node parameters are generated at random, and the output weights are computed analytically. However, even with a large number of hidden nodes, feature learning using ELM may not be efficient for natural signals due to its shallow architecture. Due to noisy signals of the smartphone sensors and high dimensional data, substantial feature engineering is required to obtain discriminant features and address the “curse-of-dimensionality”. In traditional ML approaches, dimensionality reduction and classification are two separate and independent tasks, increasing the system’s computational complexity. This research proposes a new ELM-based ensemble learning framework for human activity recognition to overcome this problem. The proposed architecture consists of two key parts: (1) Self-taught dimensionality reduction followed by classification. (2) they are bridged by “Subsampled Randomized Hadamard Transformation” (SRHT). Two different HAR datasets are used to establish the feasibility of the proposed framework. The experimental results clearly demonstrate the superiority of our method over the current state-of-the-art methods.

E-BabyNet: Enhanced Action Recognition of Infant Reaching in Unconstrained Environments.

Machine vision and artificial intelligence hold promise across healthcare applications. In this paper, we focus on the emerging research direction of infant action recognition, and we specifically consider the task of reaching which is an important developmental milestone. We develop E-babyNet, a lightweight yet effective neural-network-based framework for infant action recognition that leverages the spatial and temporal correlation of bounding boxes of infants' hands and objects to reach for to determine the onset and offset of the reaching action. E-babyNet consists of two main layers based on two LSTM and a Bidirectional LSTM (BiLSTM) model, respectively. The first layer provides a pre-evaluation of the reaching action for each hand by providing onset and offset keyframes. Then, the biLSTM model merges the previous outputs to deliver an outcome of the reaching actions detection for each frame including the reaching hand. We evaluated our approach against four other lightweight structures using a dataset comprising 5,865 annotated images resulting in 16,337 bounding boxes from 375 distinctive infant reaching actions performed while sitting by different subjects in unconstrained (home/clinic) environments. Results illustrate the effectiveness of our approach and ability to provide reliable reaching action detection and offer onset and offset keyframes with a precision of one frame. Moreover, the biLSTM layer can handle the transition between reaching actions and help reduce false detections.

SecureSense: Defending Adversarial Attack for Secure Device-Free Human Activity Recognition

Deep neural networks have empowered accurate device-free human activity recognition, which has wide applications. Deep models can extract robust features from various sensors and generalize well even in challenging situations such as data-insufficient cases. However, these systems could be vulnerable to input perturbations, i.e. adversarial attacks. We empirically demonstrate that both black-box Gaussian attacks and modern adversarial white-box attacks can render their accuracies to plummet. In this paper, we firstly point out that such phenomenon can bring severe safety hazards to device-free sensing systems, and then propose a novel learning framework, SecureSense, to defend common attacks. SecureSense aims to achieve consistent predictions regardless of whether there exists an attack on its input or not, alleviating the negative effect of distribution perturbation caused by adversarial attacks. Extensive experiments demonstrate that our proposed method can significantly enhance the model robustness of existing deep models, overcoming possible attacks. The results validate that our method works well on wireless human activity recognition and person identification systems. To the best of our knowledge, this is the first work to investigate adversarial attacks and further develop a novel defense framework for wireless human activity recognition in mobile computing research.

Federated Learning Framework for Human Activity Recognition Using Smartphones

Human activity recognition plays a vital role in various applications like healthcare, sports, and smart environments. … With the increase in the use of smart wearables and sensors, the concern for privacy has increased drastically in recognizing the activities of individuals. The legacy techniques send the data from millions of users to a central server where the processing and model generation happens. This increases the load on the server. Federated Learning is a promising method to handle sensitive and personal data in an efficient manner as well as use distributed techniques to reduce challenges that are faced in legacy approaches. Edge devices like smartphones and embedded devices extract the raw data, preprocess it, and generate a personalized model. These models are then aggregated at a central server using federated learning methods and their aggregate model is sent back to clients. Long Short-Term Memory (LSTM) is used for model generation as it can effectively capture temporal dependencies in sequential data. In order to show a real-world scenario, an AWS EC2 instance is used as a server, and laptops and smartphones are used as clients. Our results demonstrate that such methods can be adopted and can be used on a much larger scale. The performance of the global and local models is comparable to those attained by conventional methods. The utilization of federated learning shows a significant step forward in the development of human activity recognition. This approach allows for collaborative learning across multiple devices while also ensuring the privacy of users is maintained.