Using Temporal Logic and Model Checking in Automated Recognition of Human Activities for Ambient- Assisted Living Authors : Tommaso Magherini, Alessandro Fantechi, Chris D. Nugent, and Enrico Vicario IEEE TRANSACTIONS ON HUMAN-MACHINE SYSTEMS, VOL. 43, NO. 6, NOVEMBER 2013 Reporter : Nien Fu Hsiao 2015/12/05
Outline Introduction Application Scenario System Overview Experiment Conclusion 2
Introduction Increasing number of both the elderly and people suffering from long-term chronic conditions Automated activity recognition supporting the quality of life improving living experience Pervasive technologies wireless sensor networks(WSN) wired sensor networks radio-frequency identification(RFID) camera-based systems 3
Introduction Novel description-based approach formulae of propositional real-time TL (temporal logic) events detected within a smart environment Automated Recognizer of ADLs (ARA) 4
Application Scenario 5 Alice Prepare a coffee Smart Home Supervise Offer prompts
ADL Recognition Problem Require a more concrete and detailed specification of ADLs Dressing taking clothes from a wardrobe putting on overcoats Distinguish three concepts at different levels: Activity Action Observation 6
ADL Recognition Problem Activity Task undertaken by the observed subject with an intended goal that may be achieved through a sequence of different steps Action A single step possibly characterized by quantitative timing constraints that may occur in different activities Relevant Mandatory, Optional Noisy Observation A measurement taken by some device that may occur during the execution of different steps 7
Temporal Logic for ADL Propositional TL Past oriented Linear time Discrete time Real-time constraints Bounded history Real-time past linear Temporal Logic (R-pTL) 8
R-pTL Yesterday (Y), Once (O), Historically (H), and Since (S) c, d ∈ Z−, with c ≤ d ∧, ∨, and ¬ 9
ARA System 10
ARA System 11
Experiment Coffee preparation Telephone usage Take medication 12
Experiment Coffee preparation Correct preparation of a coffee Incorrect coffee preparation activity Water pouring into the cup (cpw) Addition of the ground coffee (cpc) Maximum time to prepare the coffee (tce) The delay time (tcd) 13
Experiment Correct (O [−tce, 0] cpw) ∧ (O [−tce, 0] cpc) Incorrect ((H [ −tcd, −tcd] cpc) ∧ ( ¬ (O [ −tce, 0] cpw))) ∨ ((H [ −tcd, −tcd] cpw) ∧ ( ¬ (O [ −tce, 0] cpc))) 14
Experiment Six activities addressed in the experimentation 15
Experiment in Smart Environment DANTE Activity Monitoring System DANTE marker Two cameras Dataset Creation 16
Experiment in Smart Environment Three subjects Mandatory actions Nonmandatory actions Noisy actions Randomly selected subset of the activities 12 times 17
Experiment results True positives (TP) False positives (FP) False negatives (FN) 18
Experiment results 19
Feature Selection and Activity Recognition System Using a Single Triaxial Accelerometer Authors : Piyush Gupta and Tim Dallas IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 61, NO. 6, JUNE 2014 Reporter : Nien-Fu Hsiao 2015/12/05
Outline Introduction System Components and Overview Feature Selection Activity Recognition Conclusions And Future Work 21
Introduction Falls are a major problem associated with old age It is important to develop a technology monitor gait of elderly people This paper is focused on Minimum number of sensors Minimum number of analyzing data Transitional events in ADLs Independent of the accelerometer position around the waist 22
System Components and Overview 23
System Components and Overview Belt-Clip Accelerometer AT&T Gateway MATLAB 24
Feature Selection 25
Feature Selection Six activities Walking Jumping Running Sit-to-stand/stand-to-sit Stand-to-kneel-to-stand Stationary (sitting or standing at one place) 26
Feature Selection Data signals Segmented into windows of 6 s 50% overlap 27
Feature Selection 28
Feature Selection Mean Trend and Windowed Mean Difference 6 s long acceleration series is further divided into 12 windows(0.5s each) 29
Feature Selection Variance Trend and Windowed Variance Difference 6 s long acceleration series is further divided into 12 windows(0.5s each) 30
Feature Selection DFA Coefficient alpha (α) 0.5 <0.5 >0.5 31
Feature Selection X−Z Energy Uncorrelated Cross-correlation factor 32
Feature Selection Maximum Difference Acceleration Difference between maximum and minimum acceleration experienced on each axis 33
Feature Selection Feature Selection Relief-F SFFS (sequential forward floating search) Classifier K-NN( k=10) Naive Bayes 34
Feature Selection 35
Activity Recognition Experimental Setup Seven young healthy subjects 22 and 28 years of age with no walking impairment Wear the triaxial accelerometer at their waist 36
Results 37
Results 38
Results 39
Conclusion First paper The ARA correctly recognizes the 85.4% ARA offers a formal ground to reduce the semantic gap between raw data and high-level concepts Recognition of human activities and behaviors still present several challenges 40
Conclusions Second paper System can classify different activities with high accuracy Robust system Minimum training of the users Provide least errors due to orientation and positioning offsets 41