This research explores Human Activity Recognition (HAR) for monitoring activities in smart homes using IMU wristbands, specific sampling strategies, and machine learning models to enhance accuracy. HAR systems are vital for identifying Activities of Daily Living (ADL), especially for elderly care. Data from 19 participants performing 14 activities in a smart apartment was collected using three-axis motion sensors. Preprocessing involved temporal windowing and statistical feature extraction, while dimensionality reduction was applied via PCA and SelectKBest. Four machine learning models (KNN, Random Forest, XGBoost, CatBoost) were trained with strategies like Leave-One-Subject-Out cross-validation. Longer temporal windows, such as 15 seconds, significantly improved performance. Activities like Brushing Teeth achieved an F1 score of 89%. While Chebyshev filtering had minimal impact, room-specific models and longer temporal windows demonstrated great potential for HAR in healthcare.

This paper builds on previous work to explore the feasibility of a Socially Assistive Robot (SAR)-based system for automated physical health assessments in clinical settings. Transitioning from lab experiments to real-world deployment, the TEMI robot integrates AI analytics to automate three key tests: the 30-Second Chair Stand Test (30sCST), 10-Meter Walk Test (10mWT), and Grip Strength Test (GST). Clinical evaluations with 12 healthy controls and 3 neuromuscular disease patients confirmed its reliability, matching clinician-obtained measurements. Key challenges, including network performance, video stability, and patient-specific anthropometric adjustments, were addressed. A major innovation was a locally hosted Vision Language Model (VLM) for automated grip strength data extraction, reducing errors and improving accuracy. Future work includes refining real-time data validation, expanding trials, and integrating multimodal AI for enhanced patient engagement and adaptability. This research highlights SAR-based systems as promising tools for scalable, AI-driven physical health assessments in clinical practice.

There are various traditional methods used for securing sensitive data, such as cryptography algorithms like AES-HMAC-SHA256, Twofish, and Chacha20. However, several studies showed that these cryptography algorithms suffer from security vulnerabilities. In this paper, we explore the use of a cryptography model based on a Deep Convolutional Autoencoder and we compare its performances to the cryptography algorithms. We report the results of a comparative study based on several metrics. We incorporate more nuanced metrics such as cosine similarity, entropy, Kendall and Spearman rate, and Mean Squared Error (MSE) for a comprehensive assessment of model performance and security, in addition to encryption and decryption time metrics.The results obtained are very promising. Our model performs the best on two essential metrics, entropy and MSE. We obtain a decrypted file entropy of 8.01, compared to 7.99 for the three other standard models, with a very low MSE of 0.003, compared to 105.43 for AES, which remains the most efficient compared to the other algorithms.

With the modernization of smart cities and the technological advancement of the Internet of Things, we are now reaching a point where technology can be weaved into the fabric of our cities. In this article, we tackle the challenge of using machine learning to recognize the profile of pedestrians based on their gait and silhouette with the use a thermal camera (FLIR Lepton) connected to a Raspberry Pi. We present the additional challenges faced by collecting gait data in a noisy environment, and by taking human identification a level of abstraction higher and recognizing categories of people. The far-reaching implications of such a system in terms of accessibility, inclusivity, and social participation of semi-autonomous populations are discussed. The hardware, software, and cost of the handmade prototypes used for data collection are described. In an effort to take a step toward sustainable smart cities, the possibility of powering this system using solar panels is investigated. This article aims to share the lessons learned throughout the creation and deployment of the system and to share the promising first results obtained by the team. We have reached an accuracy of 74.63% for binary age classification (adult, elder), 93.98% for mobility recognition (mobile subject, subject with reduced mobility), 85.77% for group size estimation (one subject, two subjects, or more), and 77.29% for observed gender recognition (male, female) using a 2-layer CNN without any preprocessing on very low-resolution thermal images.

In this paper, we present a method that uses the signal strength indication of RFID antennas with statistical features to perform relative positioning in a smart home. The goal of the proposed method is to enable the tracking of most objects inside a smart home in real-time, allowing activity recognition based on this tracking. This paper also introduce a new dataset of 4 100 000 RFID data collected in a real full-scale smart home setting. The dataset is available for the community. The method has an accuracy of 95.5% which is similar to previous work but require a fifth of the time to compute.

Fall detection is a major challenge for researchers. Indeed, a fall can cause injuries such as femoral neck fracture, brain hemorrhage, or skin burns, leading to significant pain. However, in some cases, trauma caused by an undetected fall can get worse with the time and conducts to painful end of life or even death. One solution is to detect falls efficiently to alert somebody (e.g., nurses) as quickly as possible. To respond to this need, we propose to detect falls in a real apartment of 40 square meters by exploiting three ultra-wideband radars and a deep neural network model. The deep neural network is composed of a convolutional neural network stacked with a long-short term memory network and a fully connected neural network to identify falls. In other words, the problem addressed in this paper is a binary classification attempting to differentiate fall and non-fall events. As it can be noticed in real cases, the falls can have different forms. Hence, the data to train and test the classification model have been generated with falls (four types) simulated by 10 participants in three locations in the apartment. Finally, the train and test stages have been achieved according to three strategies, including the leave-one-subject-out method. This latter method allows for obtaining the performances of the proposed system in a generalization context. The results are very promising since we reach almost 90% of accuracy.

Among elderly populations over the world, a high percentage of individuals are affected by physical or mental diseases, greatly influencing their quality of life. As it is a known fact that they wish to remain in their own home for as long as possible, solutions must be designed to detect these diseases automatically, limiting the reliance on human resources. To this end, our team developed a sensors platform based on infrared proximity sensors to accurately recognize basic bathroom activities such as going to the toilet and showering. This article is based on the body of scientific literature which establish evidences that activities relative to corporal hygiene are strongly correlated to health status and can be important signs of the development of eventual disorders. The system is built to be simple, affordable and highly reliable. Our experiments have shown that it can yield an F-Score of 96.94%. Also, the durations collected by our kit are approximately 6 seconds apart from the real ones; those results confirm the reliability of our kit.

One of the greatest challenge of research on smart homes is to be able to recognize the ongoing Activity of Daily Living (ADL) in real-time and make prediction on the future action of the resident. To accomplish this task, it is important to have accurate information. In this paper, we present a novel passive RFID Indoor Tracking System (ITS). The goal of this ITS was to create a simple solution from data mining that could be easily deployed in any smart environment without requiring specialized human expertise on RFID. The tracking average 80% of accuracy and show promising results for large scale deployment.

People suffering from a loss of autonomy caused by a cognitive deficit generally have to perform important daily tasks (such as cooking) using devices and appliances designed for healthy people, which do not take into consideration their cognitive impairment. Using these devices is risky and may lead to a tragedy (e.g. fire). A potential solution to this issue is to provide automated systems, which perform tasks on behalf of the patient. However, clinical studies have shown that encouraging users to maintain their autonomy greatly help to preserve health, dignity, and motivation. Therefore, we present in this paper a new smart range prototype allowing monitoring and guiding a cognitively-impaired user in the activity of preparing a meal. This new original prototype is capable of giving adapted prompting to the user in the completion of several recipes by exploiting load cells, heat sensors and electromagnetic contacts embedded in the range. We currently own a provisional patent on this new invention, and we completed a first experimental phase.

Human Activity Recognition (HAR) is a challenging problem that could enable an outstanding number of applications in pervasive computing. Many approaches have been developed to overcome this issue, but they all suffer from major drawbacks. While some use invasive sensors such as video-cameras and wearable technology, other exploit complex models to only recognize coarse-grained activities. In this paper, we propose to exploit the largely neglected spatial aspects in the smart home to recognize the activity of daily living (ADLs) of a resident in a noninvasive fashion. To do so, we designed an extension to well-known data mining algorithms that we exploit to automatically learn the models of the resident ADLs. The models are built from the retrieval of spatial patterns corresponding to the topological relationships of the smart home entities. We demonstrate the advantages of our new semi-supervised system through comprehensive experiments inside a smart home and compare the results with expert defined models of activity.

Data mining techniques have been vastly exploited recently to overcome complex problems that humans struggle to solve. Particularly, the recognition of the activity of daily living of a smart home's resident is a challenging issue that requires advanced algorithms using extensive plans' library. In this paper, we propose a novel unsupervised learning technique for the discovery of sequential pattern related to spatial relationships of objects inside a smart home. We concretely use this approach to automatically construct a library of plans. Finally, we demonstrate the efficiency with a practical activity recognition algorithm by comparing learned knowledge over expert's defined library in a real smart home.

Smart home research foresees a future where persons afflicted by a type of cognitive impairment, such as Alzheimer's disease, could pursue a longer autonomous life at home by being punctually assisted in their everyday activities. Few research teams have begun noticing that not only we need to offer appropriate services at the right time, but also that we need to adapt them to the resident profile. To do so, we must increase the accuracy and the granularity of our knowledge about the current state of environment. In this paper, we propose a new system based on the cheap and non intrusive passive RFID technology for fine grained localization of objects inside a smart home. We do so in regard for better service delivery by providing richer information about the environment.