Trung Thanh Nguyen

Multi-modal multi-view action recognition is a rapidly growing area in computer vision, with important applications in surveillance, smart homes, and assistive robotics. However, existing datasets often fail to capture real-world challenges such as distributed sensor layouts, asynchronous data streams, and limited frame-level annotations. To address these limitations, we introduce MultiSensor-Home, a novel multi-modal multi-view dataset specifically designed for realistic residential environments. It comprises 5250 untrimmed videos recorded in two distinct residential environments, Home-1 and Home-2, using five distributed RGB and audio sensor units, where each video contains multiple sequential actions along with background segments. Each frame in the recorded sequences is manually annotated with fine-grained frame-level action labels, making it, to the best of our knowledge, the first densely annotated multi-view dataset for home activity recognition. To benchmark this dataset, we further propose Action Selection Learning-guided Transformer-based Sensor Fusion (ActFusion), a unified method that jointly models temporal dynamics and cross-view correspondence. It dynamically models cross-view relationships and selects informative frames, enabling robust training under both frame-level supervision, where the start and end timings of each action are labeled, and sequence-level supervision, where only action labels are provided. To support reproducible evaluation, we establish a comprehensive benchmark with standardized training and testing protocols. Extensive experiments on MultiSensor-Home and the existing MM-Office datasets show that ActFusion consistently outperforms baseline methods across diverse scenarios. By capturing the challenges in a realistic setting, MultiSensor-Home sets a new benchmark and encourages future research on robust and generalizable action recognition methods.

Multi-view action recognition is a critical task in computer vision, with broad applications in surveillance, robotics, and video-content analysis. Traditional single-view action recognition approaches suffer from a limited field of view and occlusion, leading to incomplete understanding of actions and a higher likelihood of misclassification. Moreover, most existing methods rely on constrained environments with strong label annotations, where the onset and offset times of each action are meticulously labeled at the frame-level. However, annotating strong labels for multi-view video sequences in real-world scenarios is time consuming and labor intensive. In many cases, only a weak video-level (sequence-level) label is available, where only the action class label for the entire video sequence is provided. This limits the performance of accurate action recognition. To overcome this limitation, we propose Multi-modal Multi-view Action Selection Learning (MMASL), which integrates audio and video data to perform frame-level action recognition in large-area environments using sequence-level weak labels. The key components of MMASL include modality-specific Shared Audio Encoder and Shared Video Encoder, and an Action Selection Learning (ASL) mechanism. The encoder processes input data from multiple views by extracting and unifying features from audio and video modalities. Meanwhile, ASL dynamically selects relevant frames across views and filters out irrelevant information while focusing on critical action segments to enhance action recognition accuracy. By incorporating audio data with video data, MMASL improves recognition accuracy for visually ambiguous actions that are distinguishable through sound. Experiments in a real-world office environment using the MM-Office dataset demonstrate that MMASL outperforms state-of-the-art methods, achieving up to 8.81% improvement in mAP_C (Class-wise mean Average Precision) and 8.43% in mAP_S (Sample-wise mean Average Precision), highlighting the significance of multi-modal multi-view action recognition with ASL in real-world scenarios.

Open-vocabulary Temporal Action Detection (Open-vocab TAD) extends the detection scope of Closed-vocabulary Temporal Action Detection (Closed-vocab TAD) to unseen action classes specified by vocabularies not included in the training data, within untrimmed video. Typical Open-vocab TAD methods adopt a two-stage approach that first proposes candidate action intervals and then identifies those actions. However, errors in the first stage can affect the subsequent stage and the final detection results. Moreover, conventional methods for temporal context analyses tend to focus solely on either global or local context. Focusing solely on the global context can lead to lack of momentary detail, making it difficult to distinguish one action from another. Conversely, focusing only on the local context makes it challenging to determine the start and end timings of action intervals. To address these challenges, we introduce a one-stage approach named Hierarchical Open-vocab TAD (HOTAD), consisting of two branches: Temporal Context Analysis (TCA) and Video-Text Alignment (VTA). The former utilizes Hierarchical Encoder (HE) to fuse global and local temporal features, enabling a comprehensive capture of temporal actions, while the latter branch exploits the synergy between visual and textual modalities for precisely detecting unseen actions in the Open-vocab setting. Experiments and in-depth analysis using the widely recognized datasets THUMOS14 and ActivityNet-1.3 are performed to show the effectiveness of the proposed method. The results highlight remarkable accuracy in detecting a wide range of unseen actions. Furthermore, the proposed method significantly reduces wrong labels and localizes action instances with high precision, showcasing its robustness in complex and dynamic video settings.

Multi-modal multi-view action recognition is a rapidly growing field in computer vision, offering significant potential for applications in surveillance. However, current datasets often fail to address real-world challenges such as wide-area distributed settings, asynchronous data streams, and the lack of frame-level annotations. Furthermore, existing methods face difficulties in effectively modeling inter-view relationships and enhancing spatial feature learning. In this paper, we introduce the MultiSensor-Home dataset, a novel benchmark designed for comprehensive action recognition in home environments, and also propose the Multi-modal Multi-view Transformer-based Sensor Fusion (MultiTSF) method. The proposed MultiSensor-Home dataset features untrimmed videos captured by distributed sensors, providing high-resolution RGB and audio data along with detailed multi-view frame-level action labels. The proposed MultiTSF method leverages a Transformer-based fusion mechanism to dynamically model inter-view relationships. Furthermore, the proposed method integrates a human detection module to enhance spatial feature learning, guiding the model to prioritize frames with human activity to enhance action the recognition accuracy. Experiments on the proposed MultiSensor-Home and the existing MM-Office datasets demonstrate the superiority of MultiTSF over the state-of-the-art methods. Quantitative and qualitative results highlight the effectiveness of the proposed method in advancing real-world multi-modal multi-view action recognition.

Patients’ safety is paramount in the healthcare industry, and reducing medication errors is essential for improvement. A promising solution to this problem involves the development of automated systems capable of assisting patients in verifying their pill intake mistakes. This paper investigates a Pill-Prescription matching task that seeks to associate pills in a multi-pill photo with their corresponding names in the prescription. We specifically aim to overcome the limitations of existing pill detection methods when faced with unseen pills, a situation characteristic of zero-shot learning. We propose a novel method named Zero-PIMA (Zero-shot Pill-Prescription Matching), designed to match pill images with prescription names effectively, even for pills not included in the training dataset. Zero-PIMA is an end-to-end model that includes an object localization module to determine and extract features of pill images and a graph convolutional network to capture the spatial relationship of the pills’ text in the prescription. After that, we leverage the contrastive learning paradigm to increase the distance between mismatched pill images and pill name pairs while minimizing the distance between matched pairs. In addition, to deal with the zero-shot pill detection problem, we leverage pills’ metadata retrieved from the DrugBank database to fine-tune a pre-trained text encoder, thereby incorporating visual information about pills (e.g., shape, color) into their names, making them more informative and ultimately enhancing the pill image-name matching accuracy. Extensive experiments are conducted on our collected real-world VAIPE-PP dataset of multi-pill photos and prescriptions. Through a series of comprehensive experiments, the proposed method outperforms other methods for both seen and unseen pills in terms of mean average precision. These results indicate that the proposed method could reduce medication errors and improve patients’ safety.