Presentation is loading. Please wait.

Presentation is loading. Please wait.

Power-Accuracy Tradeoffs in Human Activity Transition Detection Prepared for DATE 2010 Dresden, Germany Jeffrey Boyd, Hari Sundaram, Aviral Shrivastava.

Similar presentations


Presentation on theme: "Power-Accuracy Tradeoffs in Human Activity Transition Detection Prepared for DATE 2010 Dresden, Germany Jeffrey Boyd, Hari Sundaram, Aviral Shrivastava."— Presentation transcript:

1 Power-Accuracy Tradeoffs in Human Activity Transition Detection Prepared for DATE 2010 Dresden, Germany Jeffrey Boyd, Hari Sundaram, Aviral Shrivastava Arizona State University

2 2 Jeffrey Boyd Arizona State University The Ideal Small Lightweight Unobtrusive Battery Life: Days, Weeks

3 3 Jeffrey Boyd Arizona State University On Low-power HW & SW: “…hardware technology has a first-order impact on the power efficiency of the system, but you've also got to have software at the top that avoids waste wherever it can. You need to avoid, for instance, anything that resembles a polling loop because that's just burning power to do nothing.” (my emphasis) -Prof. Steve Furber “A Conversation with Steve Furber,” ACM Queue, Vol. 8 No. 2, February 2010.

4 4 Jeffrey Boyd Arizona State University Tour Highlights 1.Why activity transition detection 2.Design Space 3.The great compromise 4.Design Space revisited 5.Low-power transition detection 6.Future tours

5 5 Jeffrey Boyd Arizona State University Context & Motivation Monitor patients at home Stroke rehab – Is the patient using their impaired arm? Replace surveys with objective data Classify only when you need to—at the transitions Do the minimum amount of work “Do Nothing Well” WORK

6 6 Jeffrey Boyd Arizona State University Samples, Frames, Windows, and Panes Window Size (S w ) Possible Transition Frame Size (S f ) Window Pane Sampling Frequency (F s )

7 7 Jeffrey Boyd Arizona State University Features & Temporal Resolution FeatureComputational Complexity Max O(N) Mean O(N) Min O(N) FFT O(N log N) DCT O(N log N) Haar Wavelet O(N) Daubechies Wavelet O(N) F s ={100, 50, 20, 10} Hz S f ={10, 20} samples per frame S w ={6, 8, 10, 12, 14, 16, 18, 20} seconds All combinations of accelerometer axis 4480 combinations!

8 8 Jeffrey Boyd Arizona State University Experimental Setup Five activities: Sitting, Standing, Walking, Eating, Reaching Four combinations of activities Wrist-mounted Bluetooth Connectivity 3-axis Accelerometer Processing done offline in Matlab x-axis y-axis z-axis

9 9 Jeffrey Boyd Arizona State University Sample Dataset & Evaluation Sit – Eat - Walk Peaks indicate times where the probability of transition is greatest Detect peaks, then measure: –Precision: P=Hits/(Hits + False Positives) –Recall: R=Hits/(Hits + Misses) –F-Score: F=2*P*R/(P + R) Reverse F-Score: RF = 1-F Time for each combination to process test files

10 10 Jeffrey Boyd Arizona State University Design Space & Pareto Optimal Points Faster More Accurate

11 11 Jeffrey Boyd Arizona State University Sacrifice Little, Gain Much 5% Loss 5.5x Gain

12 12 Jeffrey Boyd Arizona State University Optimal Points in Detail RFNorm. TimeSignal (axis)FeatureFreq. (Hz)Frame Size Window Size (s) 0.0360.2172xDCT1001016 0.0860.0388ymin1002018 0.1120.0359xmean1002016 0.1460.0331ymax1002014 0.1700.0330xmin1002014 0.1960.0216xmax100208 0.2700.0176xmin100206 0.3400.0172xmax100206 0.7290.0059xvariance20 10 0.7540.0056xvariance20 8 0.7750.0041xmin20 10 0.8290.0037xmean20 8 0.8780.0037zmin20 6 0.8820.0032xmean20 6 0.9380.0029xmax20 6

13 13 Jeffrey Boyd Arizona State University Scalars and Vectors

14 14 Jeffrey Boyd Arizona State University 5% Loss 5.5x Gain Summary Single-axis, simple feature Vectors are (computationally) expensive The Great Compromise 5% better accuracy or 5x battery performance Do Nothing Well

15 15 Jeffrey Boyd Arizona State University Future Tour Offerings Collect More Data! Multiple users Different Activities Train activity classifiers Build custom low-power device Implement algorithm in device firmware Reduce power by approximating features and classifiers Directed Search (for best feature and time combinations) Compare it with genetic algorithm and Monte Carlo search techniques

16 16 Jeffrey Boyd Arizona State University Fragen - Questions Contact Info: Jeffrey Boyd Jeffrey.Boyd@asu.edu Hari Sundaram Hari.Sundaram@asu.edu Aviral Shrivastava Aviral.Shrivastava@asu.edu ?


Download ppt "Power-Accuracy Tradeoffs in Human Activity Transition Detection Prepared for DATE 2010 Dresden, Germany Jeffrey Boyd, Hari Sundaram, Aviral Shrivastava."

Similar presentations


Ads by Google