Profil du membre

This work presents a real-time system for tracking multiple object in the context of meal preparation when using the Cognitive Orthosis for CoOKing (COOK). This system is called SafeCOOK. It aims to provide more capabilities to detect some dangerous situations that the current system does not consider. For example, it can locate a utensil or other kitchen object that has been left on the cooking surface of the stove while a meal is being prepared. This system uses a hybrid method based on YOLO and KCF to detect, track and drop cooking utensils as they enter and leave the cooking area, and is capable of monitoring an entire cooktop in real-time with a single camera. The software has been implemented on an embedded platform in the smart stove and has been added to it. The system produces good segmentation and tracking results at a frame rate of 1 to 4 frames per second, as demonstrated in extensive experiments using video sequences under different conditions.

Deep learning models have achieved significant success in human activity recognition, particularly in assisted living and telemonitoring. However, training these models requires substantial amounts of labeled training data, which is time-consuming and costly to acquire in real-world environments. Contrastive self-supervised learning has recently garnered attention in sensor-based activity recognition to mitigate the need for expensive large-scale data collection and annotation. Despite numerous related published papers, there remains a lack of literature reviews highlighting recent advances in contrastive self-supervised learning for sensor-based activity recognition. This paper extensively reviews 43 papers on recent contrastive self-supervised learning methods for sensor-based human activity recognition, excluding those related to video or audio sensors due to privacy concerns. First, we summarize the taxonomy of contrastive self-supervised learning, followed by a detailed description of contrastive learning models used for activity recognition and their main components. Next, we comprehensively review data augmentation methods for sensor data and commonly used benchmark datasets for activity recognition. The empirical performance comparisons of different methods are presented on benchmark datasets in linear evaluation, semi-supervised learning, and transfer learning scenarios. Through these comparisons, we derive significant insights into the selection of contrastive self-supervised models for sensor-based activity recognition. Finally, we discuss the limitations of current research and outline promising research directions for future exploration.

Deep learning models have significantly contributed to recognizing older adults’ daily activities for telemonitoring and assistance. However, recognizing human activities in real-world smart homes over the long term presents substantial challenges. Obtaining the ground truth is time-consuming and costly, yet it is crucial for training and improving deep learning models. Inspired by the impressive performance of self-supervised learning models, this paper utilizes a model based on the SimCLR framework and a self-attention mechanism for downstream human activity recognition. The model leverages the limited and intermittent labeled activities collected by the Label Older Adults’ Daily Activities (LOADA) application, which was deployed and used to acquire activity labels in the real-world, uncontrolled smart homes of three young people and two older adults for over one month. The experimental results demonstrate significant performance in activity recognition, employing semi-supervised learning with limited labels, and transfer learning scenarios where representations learned from one smart home are transferred to another. This research could inspire other human activity recognition community researchers to overcome labeling challenges for monitoring older adults in real-world scenarios.

To enable ageing in place, innovative and integrative technologies such as smart living environments may be part of the solution. Despite extensive published literature reviews on this topic, the effectiveness of smart living environments in supporting ageing in place, and in particular involving unobtrusive technologies, remains unclear. The main objective of our umbrella review was to synthesize evidence on this topic.

Deep learning models have gained prominence in human activity recognition using ambient sensors, particularly for telemonitoring older adults’ daily activities in real-world scenarios. However, collecting large volumes of annotated sensor data presents a formidable challenge, given the time-consuming and costly nature of traditional manual annotation methods, especially for extensive projects. In response to this challenge, we propose a novel AttCLHAR model rooted in the self-supervised learning framework SimCLR and augmented with a self-attention mechanism. This model is designed for human activity recognition utilizing ambient sensor data, tailored explicitly for scenarios with limited or no annotations. AttCLHAR encompasses unsupervised pre-training and fine-tuning phases, sharing a common encoder module with two convolutional layers and a long short-term memory (LSTM) layer. The output is further connected to a self-attention layer, allowing the model to selectively focus on different input sequence segments. The incorporation of sharpness-aware minimization (SAM) aims to enhance model generalization by penalizing loss sharpness. The pre-training phase focuses on learning representative features from abundant unlabeled data, capturing both spatial and temporal dependencies in the sensor data. It facilitates the extraction of informative features for subsequent fine-tuning tasks. We extensively evaluated the AttCLHAR model using three CASAS smart home datasets (Aruba-1, Aruba-2, and Milan). We compared its performance against the SimCLR framework, SimCLR with SAM, and SimCLR with the self-attention layer. The experimental results demonstrate the superior performance of our approach, especially in semi-supervised and transfer learning scenarios. It outperforms existing models, marking a significant advancement in using self-supervised learning to extract valuable insights from unlabeled ambient sensor data in real-world environments.

Within large and growing human communities where interactions occur, trust is a key factor to consider. Computational trust models have then been widely studied since the 2000s targeting items ratings (e.g. in e-commerce) or M2M (e.g. in IoT network). Among these models, EigenTrust is today one of the most popular and studied ones. It provides a global reputation calculation and is efficient in distributed networks, but not fully satisfactory for human interactions. On the opposite, the Bi-lattice model is well suited for human networks interactions such as solidarity networks and/or human services exchange networks but is limited to local trust results. In this paper, we propose a new aggregator that extends the Bi-lattice model to enable a global reputation calculation. This new aggregator discovers the trust links from the member whose score is to be evaluated to every other members he is connected to on the trust network. It then computes the global reputation of this member based on these trust links. Furthermore, it enables a lightweight approach, as it is able to compute a global reputation based only on a partial knowledge of the trust network. Throughout the paper, the proposed aggregator is presented, evaluated and compared to Eigentrust to show its effectiveness.

This work presents a real-time system for tracking multiple object in the context of meal preparation when using the Cognitive Orthosis for CoOKing (COOK). This system is called SafeCOOK. It aims to provide more capabilities to detect some dangerous situations that the current system does not consider. For example, it can locate a utensil or other kitchen object that has been left on the cooking surface of the stove while a meal is being prepared. This system uses a hybrid method based on YOLO and KCF to detect, track and drop cooking utensils as they enter and leave the cooking area, and is capable of monitoring an entire cooktop in real-time with a single camera. The software has been implemented on an embedded platform in the smart stove and has been added to it. The system produces good segmentation and tracking results at a frame rate of 1 to 4 frames per second, as demonstrated in extensive experiments using video sequences under different conditions.

Human activity recognition (HAR) using ambient sensors has emerged as a promising approach to telemonitoring daily activities and enhancing the elderly quality of life. Deep learning models have demonstrated competitive performance in HAR on real-world datasets. However, acquiring large amounts of annotated sensor data for extracting robust features is costly and time-consuming. To overcome this limitation, we propose a novel model based on the self-supervised learning framework, SimCLR, for daily activity recognition using ambient sensor data. The core component of the model is the encoder module, which consists of two convolutional layers followed by a long short-term memory (LSTM) layer. This architecture allows the model to capture both spatial and temporal dependencies in the sensor data, enabling the extraction of informative features for downstream tasks. Through extensive experiments on three CASAS smart home datasets (Aruba-1, Aruba-2, and Milan), we showcase the superior performance of the model in semi-supervised learning and transfer learning scenarios, surpassing state-of-the-art approaches. The findings highlight the potential of self-supervised learning in extracting valuable information from unlabeled sensor data, reducing costly annotation efforts for real-world HAR applications.

In Quebec, home care administrators are increasingly open to using Ambient Assisted Living (AAL) technologies as part of services to better support care recipient with a major loss of autonomy. However, little information is available about how these technologies are integrated into clinical practice.

Remote monitoring uses smart home features to promote aging in place by preventing emergencies and increasing the quality of life of older adults. However, traditional reports, data, and graphs produced by remote monitoring technologies are not well suited to older adults’ needs. Thus, the complexity for older adults to use and interpret reports can lead to usability and adoption issues. The goals of this study were 1) to incorporate ludic-based design principles into an application that provides older adults with an alternative way to interact with information about their Activities of Daily Living (ADL), and 2) involve older adults in creating new ludic interfaces that address usability and reduce adoption issues. This ambient assistive technology offers older adults the opportunity, through its interface, to promote curiosity and exploration, the pursuit of non-external goals, and openness about the user’s routine and lifestyle. By using an iterative, Human-Centered, co-design approach in 4 workshops with older adults (N = 7), we combine older adults’ needs with ludic elements to propose a new user experience.