Search

Human Activity Recognition (HAR) has emerged as a vital measure of quality of life, holding significant implications for human health. The need for effective mobility monitoring in diverse settings, both indoors and outdoors, necessitates the development of scientific and technological tools. To broaden accessibility, wearable devices like smartphones and smartwatches are commonly employed, leveraging sensor data analysis for valuable insights into user activities. HAR, treated as a classification task, involves training a classifier using sensor data and corresponding activity labels. The classifier aims to automatically recognize and classify various activities in future instances. However, the performance of a generic HAR model trained on a diverse population diminishes significantly when applied to a specific user due to inter-subject variability, encompassing variations in activity patterns, behavioral status, gait, and posture. To address this challenge, we propose an innovative auto-supervised domain adaptation approach based on particle filter theory, aiming to automatically construct a personalized HAR classifier. Our approach integrates multiple steps inspired by the particle filter formalism, enabling iterative approximation of the target distribution through temporal samples to personalize the HAR model for the specific user. In extensive experiments on public HAR datasets, we emphasize the critical role of personalization when deploying an HAR classifier for a new user. The results demonstrate that our framework significantly enhances the accuracy of HAR for new users compared to a non-personalized model, achieving an average improvement of 50% across most datasets. Furthermore, we implement our personalized HAR model on an embedded wearable device, enhancing its accessibility for real-world applications. [ABSTRACT FROM AUTHOR]

Terrestrial Laser Scanning (TLS) systems have gained increasing popularity in the forestry domain and are today widely used for the automatic measurement of forest inventory attributes. Nevertheless, to the best of our knowledge the problem of tree species recognition from TLS data has received very little attention from the scientific community. It is in this context that we present a novel Computer-Aided Tree Species Identification method based on 3D bark texture analysis. The novelty of our approach resides in the following three key points: (1) 3D salient regions extraction using a new morphological segmentation method that we have called Burst Wind Segmentation, (2) the extraction and pre-annotation of a collection of typical 3D bark patterns, known as scars, from each of the tree species. The pre-annotated scars are stored in a dictionary that we have called ScarBook and they are used as a reference for the comparison of the 3D salient segmented regions, (3) a wide variety of advanced shape, saliency, curvature and roughness features are extracted from the 3D salient segmented regions. To study the performance of our method, an experiment has been carried out on a dataset composed of 969 patches which correspond to 30 cm long segments of the trunk at breast height. Six species among the most dominant species in European forests have been tested with patches of different diameter at breast height values so as to study the identification accuracy with respect to age. The results obtained are very encouraging and promising and they confirm the possibility of identifying tree species using TLS data. [ABSTRACT FROM AUTHOR]

Tactile maps, and even more audio-tactile maps (Brock et al., 2012) are useful tools for visually impaired people to grasp the geographic space that they cannot experience with their sight. Due to low precision of touch compared to sight, these tactile maps have to be very simplified and generalised compared to their counterparts designed for people with regular sight (Touya et al., 2019). This is why it is very complex for orientation and mobility trainers to design and create enough of these maps for their needs. The orientation and mobility trainers we interviewed claim that they need a tool for non-expert GIS users to fasten the time to design and print a new audio-tactile map. Some simple tools exist, such as TouchMapper1, or Hapticke2, which allow the generation of a tactile map anywhere on earth from OpenStreetMap data. But these tools fail to generate maps that are generalised enough (Touya et al., 2019). This is why we started the design of the MULTIMAGE platform (MULtimodal Tactile and Interactive Map GEnerator). The MULTIMAGE platform proposes a workflow with three steps to generate a 3D model ready to be printed from OpenStreetMap data (Figure 1). The first step is just the selection of the extent, the scale, and the size of the map, and then an automated extraction of the corresponding raw data from OpenStreetMap. For now, our focus is on maps of crossroads at large scales, because crossroads are one of the main issues during the training of autonomous mobility for visually impaired people. The extent of the map is simply selected by drawing a bounding box on a web map. Then, the second step of the workflow is the data enrichment, which enables the use of automated enrichment scripts, as well as interactive actions to correct potentially incomplete or incorrect OpenStreetMap data. The third step is the design of the map where stylisation and map generalisation tasks can be carried out, either automatically or interactively. Then, the map is exported as an extended STL file to be 3D-printed, with information to print buttons of audio description. We also plan the platform to be able to export files adapted to swell paper printers, as swell paper (Figure 2) can be used on top of tablets to make interactive audio-tactile devices (Brok et al., 2012). The interactive actions are possible using the map interface presented in Figure 3. As a classical web GIS tool, the edition of the geometry and the semantics of the OSM data will be possible in the second step, and the edition of style as well as manual generalisation operations will be possible in the third step. Regarding the automatic scripts, several are already implemented in the platform. For instance, there is a script to segment and enrich the crossroad with important features such as traffic islands (Favreau & Kalsron, 2022), and a script to generate styles adapted to an audiotactile map (Jiang et al., 2021). The backend, i.e. the server side of the platform is implemented in Python, using open source libraries such as GeoDjango and Geopandas. The web interface is developed in Javascript, based on the OpenLayers library. Data are stored using PostGIS, and the Web Feature Services displayed in the interface are generated with Geoserver. [ABSTRACT FROM AUTHOR]