Download presentation
Presentation is loading. Please wait.
1
WP3 INERTIA Local Control and Automation Hub
Task 3.4 Multi-Sensorial Activity Flow Detection & Modelling in controlled environments CERTH-ITI
2
Task Progress / Main Achievements
Occupancy Extraction Multi-sensorial occupancy extraction algorithm based on Hidden Markov Model (HMM) Number of occupants as the HMM hidden state, estimated periodically Pre-Processing Component: Handles sensor events, prepares the input for the HMM, monitors time-related functions
3
Task Progress / Main Achievements
Inputs: - Installed sensor events from monitored spaces - HMM model Output: Estimated occupancy per monitored space Publish an Occupancy_Change event if occupancy is changed in one space
4
Task Progress / Main Achievements
Preliminary indicative results with simulated input data simulated sensor data for a small office 1 double BEAM sensor at the door 1 PIR (motion) sensor Observations BEAM count number (0-3) PIR activation (0,1)
5
Task Progress / Main Achievements
Acoustic Sensor Tests Logging noise levels in kitchen Investigate correlation between noise level and occupants density occupancy detection (empty-occupied) is quite accurate Occupancy density extraction not so accurate Around 180th sec, people arrived for lunch (We observe intense variations in noise levels)
6
Task Progress / Main Achievements
Integration of sensors using Arduino Pair of Pressure mats (binary input) at doors- to count occupants entering/leaving space Arduino Uno R3 with WiFi Shield for wireless data transmission Data collection and storing to DB [SensorID, DateTime, Value] Double-beam sensors integration in progress in a similar manner Double Beam Sensors Arduino Pressure Mats
7
Task Progress / Main Achievements
Occupancy Prediction Investigation of alternative methods and mathematical models Prototype development extending functionalities and implementing additional algorithmic approaches based on the Semi-Markov Model Evaluation of the performance of the various algorithmic methods implemented so far using real occupancy data (2 weeks) from Kinect cameras collected from pilot sites Evaluation metrics: total NRMSE (general) and weighted NRMSE (more relevant to the INERTIA concept) Results so far show that INERTIA prediction in most cases is far better than historical average and respective Open Reference Models
8
Task Progress / Main Achievements
Data test installations for data gathering Kinect Cameras Cameras have already been installed in CERTH premises (developers office, kitchen, rest area) to acquire more data for testing and evaluation purposes Data from Kinect cameras will be used in some cases as ground truth for the validation of the other occupancy extraction methods Combinations of occupancy sensors to be installed: Pressure mats / Double beam + PIR (available through the alarm system) Pressure mats / Double beam + PIR + Acoustic + CO2 PIR + Acoustic PIR + Acoustic + Electric appliances usage detection through smart plugs
9
Allocation of work among partners
Interfacing with other components TUK: Initial discussions for interfacing with IAM Initial example services have been implemented for testing Define XSD schemas for all services needed CNet: Provide access to LinkSmart Define way to publish and subscribe to events Store real-time values (occupancy extraction, occupancy prediction??)
10
Critical issues towards Task’s completion
Pending / foreseen issues Real-time occupancy extraction changes stored to LinkSmart Event Manager Other components will request current occupancy per space from LinkSmart (not IAM) LinkSmart will automatically store occupancy to IAM (to be used for historical purposes) Raw sensor events from a double beams sensor may be too fast to be logged with accurate time. Perhaps it is preferable to process single raw events at the lowest level (Arduino) and directly get the entry/exit event.
11
Critical issues towards Task’s completion
Pending / foreseen issues Occupancy prediction Define prediction scenarios (e.g. normal operation, demand response) Will be produced by Occupancy Component automatically on a regular basis (e.g. per 1 to 15 minutes?) Will also be asked by the Holistic Flexibility Component and passed directly to it? Event for prediction change passed to Event Manager – actual data stored in LinkSmart or IAM? Exported XML will consist of occupancy prediction per space / zone for a few hours after current time (e.g. 2 hours) / whole remaining day / next day? values per minute
12
Deliverable D3.3 Deliverable D3.3 Ambient User Interfaces, User Behavioural Profiling and Activity Flow Framework Combined with T3.5 Table of Contents available Initial version available with inputs from HYPERTECH on “UI and User Profiling sections” CERTH will fully provide “Activity Flow Framework” section Planning: Intermediate version by end of August Peer review version by mid of September
13
Deliverable D3.3 TOC – Activity Flow Framework section
1. Occupancy Extraction Framework 1.1 Introduction 1.2 Analysis of available occupancy detection systems 1.3 Selection of optimal multi-sensor systems in the scope of INERTIA 1.3.1 Identification of different tertiary building rooms and zones 1.3.2 Methodology for testing different multi-sensor installations per identified rooms/zones 1.3.3 Prototype implementation for evaluation of different occupancy extraction algorithms 1.4 Analysis of CERTH pilot test results 1.5 Conclusions on best fitted business scenarios in the scope of INERTIA 2. Activity Flow Modelling and Occupancy Prediction Framework 2.1 Introduction 2.2 Available occupancy prediction methods 2.3 Selected methods and algorithmic approaches in the scope of INERTIA 2.3.1 Implementation of selected approaches 2.3.2 Incorporation of scheduling information/ Open Reference Models 2.4 Experimental results for framework evaluation 2.5 Conclusions 3. ANNEX: Sensor detailed description
14
Planning for next efforts
Occupancy Extraction Sensors foreseen for pilot installation (i.e. PIR, Pressure mats, Double Beam Sensors, Acoustic, CO2 ) to be implemented, tested and fully supported be prototype Initial accuracy validation per sensor combination Improve algorithm’s sensor modelling methods based on results Initial progress on Individual Extraction using RFID Occupancy Prediction Incorporate scheduling information (typical and dynamic) Acquire more data from pilot premises (from Kinect cameras and other sensors) towards better assessment of the different approaches Use more specific and INERTIA oriented evaluation metrics Add individual prediction support (based on RFID extraction)
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.