Hips do lie! A position-aware mobile fall detection system Christian - PowerPoint PPT Presentation

Hips do lie! A position-aware mobile fall detection system Christian Krupitzer, Timo Sztyler, Janick Edinger, Martin Breitbach, Heiner Stuckenschmidt, and Christian Becker

Proposition So be wise and keep on Reading the signs of my body And I'm on tonight you know my hips don't lie. S HAKIRA : “H IPS D ON ’ T L IE ” (2005) Hips Do Lie! A Position-Aware Mobile Fall Detection System C. Krupitzer, T. Sztyler, J. Edinger, M. Breitbach, H. Stuckenschmidt, and C. Becker

Alternative Proposition Observation: Hips might not be the most reliable source of information Hips Do Lie! A Position-Aware Mobile Fall Detection System 3 C. Krupitzer, T. Sztyler, J. Edinger, M. Breitbach, H. Stuckenschmidt, and C. Becker

Research Question ? How can a mobile fall detection system react to changes in the sensor position? Hips Do Lie! A Position-Aware Mobile Fall Detection System C. Krupitzer, T. Sztyler, J. Edinger, M. Breitbach, H. Stuckenschmidt, and C. Becker

What to Expect from this Presentation? Fall Detection in a Nutshell Self-Adaptive Fall Detection Implemen- tation Case Studies Future Work Hips Do Lie! A Position-Aware Mobile Fall Detection System 5 C. Krupitzer, T. Sztyler, J. Edinger, M. Breitbach, H. Stuckenschmidt, and C. Becker

Fall Detection Systems Vision-based Ambient-based Wearables Mubashir, Shao & Seed: A survey on fall detection: Principles and approaches. In: Neurocomputing, Volume 100, 2013, Pages 144-152. Hips Do Lie! A Position-Aware Mobile Fall Detection System 6 C. Krupitzer, T. Sztyler, J. Edinger, M. Breitbach, H. Stuckenschmidt, and C. Becker

Positions of Wearables for Fall Detection Fall Jump Sit Walk Bend Run Sit Walk Get up Walk Jump Head Chest Wrist Waist Ankle Hip Hips Do Lie! A position-aware mobile fall detection system 7 C. Krupitzer, T. Sztyler, J. Edinger, M. Breitbach, H. Stuckenschmidt and C. Becker

Self-Adaptive Fall Detection DP FD FA DA DP FD FA DA Data Data Fall Fall alert acquisition preprocessing detection Magn. Analyzer Planner Gyro X Notification Monitor Executor Y Z Acc. Acc. Caregivers Data Patient Sensors Adaptation Logic Hips Do Lie! A Position-Aware Mobile Fall Detection System 8 C. Krupitzer, T. Sztyler, J. Edinger, M. Breitbach, H. Stuckenschmidt, and C. Becker

Self-Adaptive Fall Detection DP FD FA DA DP FD FA DA Data Data Fall Fall alert acquisition preprocessing detection FD Analyzer Planner FA DA DP X Notification Monitor Executor Y Z Acc. Acc. Caregivers Data Patient Sensors Adaptation Logic Hips Do Lie! A Position-Aware Mobile Fall Detection System 9 C. Krupitzer, T. Sztyler, J. Edinger, M. Breitbach, H. Stuckenschmidt, and C. Becker

Implementation: Adaptation Logic Fall Analyzer Planner Notification ? Fall Trigger Handling Acc. Data (x,y,z) Fall Alert Acc. Data: (x,y,z) Notifi- Monitor Executor cation Hips Do Lie! A Position-Aware Mobile Fall Detection System 10 C. Krupitzer, T. Sztyler, J. Edinger, M. Breitbach, H. Stuckenschmidt, and C. Becker

Implementation: Adaptation Logic Fall Analyzer Planner Notification Window Manager t 1 : (x 1 ,y 1 ,z 1 ) w 1 Feature 1 Feature 2 ... ... Fall t n : (x n ,y n ,z n ) Trigger Handling ... w 2 Feature 1 Feature 2 ... t n+x : (x n+x ,y n+x ,z n+x ) ... ... ... ... Acc. Data: (x,y,z) Notifi- Monitor Executor cation Hips Do Lie! A Position-Aware Mobile Fall Detection System 11 C. Krupitzer, T. Sztyler, J. Edinger, M. Breitbach, H. Stuckenschmidt, and C. Becker

Implementation: Adaptation Logic Fall Analyzer Planner Notification Window Manager Fall Trigger versus Handling Event-based approach Continuous approach Acc. Data: (x,y,z) Notifi- Monitor Executor cation Hips Do Lie! A Position-Aware Mobile Fall Detection System 12 C. Krupitzer, T. Sztyler, J. Edinger, M. Breitbach, H. Stuckenschmidt, and C. Becker

Implementation: Adaptation Logic Fall Analyzer Planner Notification Window Manager Fall Trigger versus Handling Event-based approach Continuous approach Acc. Data: (x,y,z) Notifi- Monitor Executor cation Hips Do Lie! A Position-Aware Mobile Fall Detection System 13 C. Krupitzer, T. Sztyler, J. Edinger, M. Breitbach, H. Stuckenschmidt, and C. Becker

Self-Adaptive Fall Detection FD FA DP DA Data Data Fall Fall alert acquisition preprocessing detection FD Analyzer Planner FA DA DP X Notification Monitor Executor Y Z Acc. Acc. Caregivers Data Patient Sensors Adaptation Logic Hips Do Lie! A Position-Aware Mobile Fall Detection System 14 C. Krupitzer, T. Sztyler, J. Edinger, M. Breitbach, H. Stuckenschmidt, and C. Becker

Self-Improving Fall Detection X-Means Sztyler 2016 Index (BMI) Body Mass Age A 1 A 2 … A n Position Subject Analyzer Data Clusterer Algorithms Analyzer Planner Instructions Subject Analyzer Algorithm Chooser Adaptation Executor Data Monitor Monitor Executor Algorithm Adaptation Logic Self-Improvement Module Optimizer Hips Do Lie! A Position-Aware Mobile Fall Detection System 15 C. Krupitzer, T. Sztyler, J. Edinger, M. Breitbach, H. Stuckenschmidt, and C. Becker

Evaluation Setting Algorithms: • Propositions: Classification-based: SVM, ANN, Random Forest, k-NN, 1. Origin of data matters 2. Position matters J48 Decision Tree • 3. Self-Improvement matters Threshold-based (params learned) • Supports deep learning, but dataset is too small Data: #32 #10 #30 #23 Training of models / evaluation: Leave-one-subject-out approach Training Testing Set Set MMSys UMA UniMiB SisFall Hips Do Lie! A Position-Aware Mobile Fall Detection System 16 C. Krupitzer, T. Sztyler, J. Edinger, M. Breitbach, H. Stuckenschmidt, and C. Becker

Evaluation Overview Experiment 1: Cross dataset analysis Experiment 2: Position-aware fall detection Experiment 3: Self-improving fall detection Hips Do Lie! A Position-Aware Mobile Fall Detection System 17 C. Krupitzer, T. Sztyler, J. Edinger, M. Breitbach, H. Stuckenschmidt, and C. Becker

Experiment 1: Cross-Datasets Fall Detection Trained across datasets Tested on Trained on same dataset 0,80 0,73 0,71 0,68 0,63 F2-Scores ANN Trained across 0,60 0,54 0,52 0,48 datasets 0,45 0,40 Trained on same datasets 0,20 0,00 MMSys UMA UniMiB SisFall • Proposition 1: Origin of data matters • The results indicate issues in labeling and setup of capturing data Hips Do Lie! A Position-Aware Mobile Fall Detection System 18 C. Krupitzer, T. Sztyler, J. Edinger, M. Breitbach, H. Stuckenschmidt, and C. Becker

Experiment 2: Position-Aware Fall Detection Trained on one position Tested on all positions 0,82 0,80 0,80 0,76 0,73 0,80 Model trained 0,69 0,65 0,61 0,57 on Chest F1-Scores 0,60 Model trained 0,40 on Hip 0,20 Model trained on Waist 0,00 Tested on Chest Tested on Waist Tested on Hip • Proposition 2: Position matters • The results indicate that hips performed worst Hips Do Lie! A Position-Aware Mobile Fall Detection System 19 C. Krupitzer, T. Sztyler, J. Edinger, M. Breitbach, H. Stuckenschmidt, and C. Becker

Experiment 3: Self-Improving Fall Detection Falls Raw Data Position Hips Do Lie! A position-aware mobile fall detection system 20 C. Krupitzer, T. Sztyler, J. Edinger, M. Breitbach, H. Stuckenschmidt and C. Becker

Experiment 3: Self-Improving Fall Detection Falls Raw Data Position Hips Do Lie! A position-aware mobile fall detection system 21 C. Krupitzer, T. Sztyler, J. Edinger, M. Breitbach, H. Stuckenschmidt and C. Becker

Experiment 3: Self-Improving Fall Detection Falls Raw Data Position Hips Do Lie! A position-aware mobile fall detection system 22 C. Krupitzer, T. Sztyler, J. Edinger, M. Breitbach, H. Stuckenschmidt and C. Becker

Experiment 3: Self-Improving Fall Detection Falls Raw Data Position Hips Do Lie! A position-aware mobile fall detection system 23 C. Krupitzer, T. Sztyler, J. Edinger, M. Breitbach, H. Stuckenschmidt and C. Becker

Experiment 3: Self-Improving Fall Detection Falls Raw Data Position Hips Do Lie! A position-aware mobile fall detection system 24 C. Krupitzer, T. Sztyler, J. Edinger, M. Breitbach, H. Stuckenschmidt and C. Becker

Experiment 3: Self-Improving Fall Detection Detected falls: ANN Detected falls: ANN J48 J48 KNN KNN Random Forrest Random Forest ALM Self-Improving XYZ acceleration data Position Chest (Ground truth) Hip Chest (Classified) Hip • Recall: 0.93 - Any falls missed? • No: We missed fall windows, but for each fall we detected at least one window! Hips Do Lie! A position-aware mobile fall detection system 25 C. Krupitzer, T. Sztyler, J. Edinger, M. Breitbach, H. Stuckenschmidt and C. Becker

Experiment 3: Self-Improving Fall Detection Detected falls: ANN Detected falls: ANN J48 J48 KNN KNN Random Forrest Random Forest ALM Self-Improving XYZ acceleration data Position Chest (Ground truth) Hip Chest (Classified) Hip • Proposition 3: Self-improvement matters • Precision is improved  decreases false positives Hips Do Lie! A position-aware mobile fall detection system 26 C. Krupitzer, T. Sztyler, J. Edinger, M. Breitbach, H. Stuckenschmidt and C. Becker

Current State X Y Z Acc. Hips Do Lie! A Position-Aware Mobile Fall Detection System 27 C. Krupitzer, T. Sztyler, J. Edinger, M. Breitbach, H. Stuckenschmidt, and C. Becker