2000 Energy Saving 1800 Screen Off Prevew with 1600 Screen On Minimum Phone Lifting Arm Frame Rate Rotating 1400 Turn off Camera Mationless Adjust Screen Walking Brizhtness based Up Preview with Fine 1200 Turn Offnior on Ambient Light uning 1000 without Low the 176*144320*240720°4809607201280720 Shooting for a Laving Arn Brightness to 5 Down Shoot Long Time ame Rate Fig.13. Energy Consumption with Different Resolution of Sony LT26w Maximum Medium Minimum Fig.15. Energy Saving Schemes Corresponding to Different States IV.SYSTEM EVALUATION We implement our system on Samsung GT-19250 smart- phone running on Google's Android platform.The version of the Android system is 4.4.2.We use Monsoon power monitor to measure the power consumption of the phone. (a)Different Frame Rate of Preview (b)Different Brightness of Screen Fig.14.Energy Consumption of Different Brightness and Preview Frame Rates A.Impact of Sensors'Sampling Rate 1)Energy Consumption of Sensors with Various Sampling Rates:The sensors'maximum sampling rate of Samsung GT- of Android mechanism,the phone adaptively chooses the 19250 is 100 Hz.We vary the sampling rate in twenty levels suitable preview frame rate in the range by itself.Therefore with step of 5.Energy consumption of sensors are different the energy consumption of last two situations are same. with various sampling rates as shown in Figure 16(a).We Observation of Screen.Brightness of screen is related to observe that power consumed is relatively big after 25 Hz. the energy consumption shown in Figures 14(b).The range of 2)Energy Consumption of Calculation of Sensors'Data: brightness is 0-255.0 stands for darkest and 255 for brightest. Energy consumption of calculation is related to the data size Once the brightness drops,the energy consumption decreases. With the sampling rate increasing,the energy also increases. 2)Energy Saving Scheme:For obtained states,we apply We observe that the power of calculation of sliding window is corresponding energy saving strategies,as shown in Figure 15 only 1.5 mW.SVM model consumes about 4.34 mW as it is only used to do prediction with a trained model.The power If obtained state is in body level,the screen and the camera of FFT is about 122 mW.The energy of other calculation can will be turned off as the user doesn't need to look at the screen. be ignored.Therefore,only energy of FFT is needed to be Further more,if the states always belong to body level for a considered.But compared to the power of camera which is long time(15 minutes is chosen in our implement),the sensors about 1500 mW,it is unobservable. will be turned off until the camera software is used again. 3)Activity Recognition Accuracies with Various Sampling If obtained state is in arm level,the screen is turned on Rates:We use six sampling rates,which are 2 samples/second. and its brightness is adjusted based on ambient light as lifting 5 samples/second,10 samples/second,20 samples/second,50 up your arm.The brightness will be declined to 5 as laying samples/second and 100 samples/second,to evaluate the im- down your arm.The brightness is set to five levels according pact of sampling rate of sensors on the performance of recog- to different environment shown in Table II. nition accuracy.In Figures 16(b),the accuracies of actions in body level are showed.For motionlessness,the accuracy has TABLE II. THE BRIGHTNESS OF SCREEN IN DIFFERENT ENVIRONMENT no relationship with the sampling rate.For walk,the accuracy is low with rate of 2 Hz and bigger than 90%with the other Number Environment Ambient Light (SI lux) of Screen five rates.For jog,the accuracy is bigger than 90%at the rate Day.outdoor,sunny >6000 180 of 10 Hz.In Figures 16(c),the accuracies of actions in arm Day,outdoor,cloudy 5006000T 130 level are showed.We can find out that the accuracy of 20 Hz Day,indoor,no lamp 100300 80 is 100%.In Figures 16(d).the accuracies of actions in wrist Night,outdoor,street lamp 100 55 Night,indoor,lamp 300500 105 level are showed and it is 100%when the sample rate is 100 Hz. If obtained state is in wrist level,the camera will be turned on and stay in preview mode.When you rotate the phone, B.Trade off between Energy Consumption and Recognition Accuracy camera is set to work with smallest frame rate supported by the phobe.In fine-tuning state,camera works with increased From Figures 16(b)(c)(d).the accuracy can be 100%with frame rate (median value is used).And in shooting state,all 100 Hz.But it not energy efficiency as shown in Figure the indexes will return to normal.All the parameters can be 16(a).Therefore,we need make a trade off between energy changed by the user if the parameters do not fit them consumption and recognition accuracy.176*144 320*240 720*480 960*720 1280*720 Power (mW) 1000 1200 1400 1600 1800 2000 Fig. 13. Energy Consumption with Different Resolution of Sony LT26w 15-15 15-30 24-30 Power of Preview (mW) 660 680 700 720 740 760 780 800 (a) Different Frame Rate of Preview Brightness of Screen 255 230 205 180 155 130 105 80 55 30 5 Power of Screen(mW) 400 500 600 700 800 900 1000 1100 1200 1300 1400 (b) Different Brightness of Screen Fig. 14. Energy Consumption of Different Brightness and Preview Frame Rates of Android mechanism, the phone adaptively chooses the suitable preview frame rate in the range by itself. Therefore the energy consumption of last two situations are same. Observation of Screen. Brightness of screen is related to the energy consumption shown in Figures 14(b). The range of brightness is 0-255. 0 stands for darkest and 255 for brightest. Once the brightness drops, the energy consumption decreases. 2) Energy Saving Scheme: For obtained states, we apply corresponding energy saving strategies, as shown in Figure 15. If obtained state is in body level, the screen and the camera will be turned off as the user doesn’t need to look at the screen. Further more, if the states always belong to body level for a long time (15 minutes is chosen in our implement), the sensors will be turned off until the camera software is used again. If obtained state is in arm level, the screen is turned on and its brightness is adjusted based on ambient light as lifting up your arm. The brightness will be declined to 5 as laying down your arm. The brightness is set to five levels according to different environment shown in Table II. TABLE II. THE BRIGHTNESS OF SCREEN IN DIFFERENT ENVIRONMENT Number Environment Ambient Brightness Light (SI lux) of Screen 1 Day, outdoor, sunny >6000 180 2 Day, outdoor, cloudy 500∼6000 130 3 Day, indoor, no lamp 100∼300 80 4 Night, outdoor, street lamp <100 55 5 Night, indoor, lamp 300∼500 105 If obtained state is in wrist level, the camera will be turned on and stay in preview mode. When you rotate the phone, camera is set to work with smallest frame rate supported by the phobe. In fine-tuning state, camera works with increased frame rate (median value is used). And in shooting state, all the indexes will return to normal. All the parameters can be changed by the user if the parameters do not fit them. Maximum Medium Minimum Screen Off Turn Off Sensors without Shooting for a Long Time Screen On Adjust Screen Brightness based on Ambient Light Preview with Minimum Frame Rate Turn off Camera Energy Saving Motionless Walking, Jogging, Low the Brightness to 5 Lifting Arm Up Laying Arm Down Phone Rotating Preview with Median Frame Rate Fine￾tuning Preview with Normal Frame Rate Shooting Fig. 15. Energy Saving Schemes Corresponding to Different States IV. SYSTEM EVALUATION We implement our system on Samsung GT-I9250 smart￾phone running on Google’s Android platform. The version of the Android system is 4.4.2. We use Monsoon power monitor to measure the power consumption of the phone. A. Impact of Sensors’ Sampling Rate 1) Energy Consumption of Sensors with Various Sampling Rates: The sensors’ maximum sampling rate of Samsung GT￾I9250 is 100 Hz. We vary the sampling rate in twenty levels with step of 5. Energy consumption of sensors are different with various sampling rates as shown in Figure 16(a). We observe that power consumed is relatively big after 25 Hz. 2) Energy Consumption of Calculation of Sensors’ Data: Energy consumption of calculation is related to the data size. With the sampling rate increasing, the energy also increases. We observe that the power of calculation of sliding window is only 1.5 mW. SVM model consumes about 4.34 mW as it is only used to do prediction with a trained model. The power of FFT is about 122 mW. The energy of other calculation can be ignored. Therefore, only energy of FFT is needed to be considered. But compared to the power of camera which is about 1500 mW, it is unobservable. 3) Activity Recognition Accuracies with Various Sampling Rates: We use six sampling rates, which are 2 samples/second, 5 samples/second, 10 samples/second, 20 samples/second, 50 samples/second and 100 samples/second, to evaluate the im￾pact of sampling rate of sensors on the performance of recog￾nition accuracy. In Figures 16(b), the accuracies of actions in body level are showed. For motionlessness, the accuracy has no relationship with the sampling rate. For walk, the accuracy is low with rate of 2 Hz and bigger than 90% with the other five rates. For jog, the accuracy is bigger than 90% at the rate of 10 Hz. In Figures 16(c), the accuracies of actions in arm level are showed. We can find out that the accuracy of 20 Hz is 100%. In Figures 16(d), the accuracies of actions in wrist level are showed and it is 100% when the sample rate is 100 Hz. B. Trade off between Energy Consumption and Recognition Accuracy From Figures 16(b)(c)(d), the accuracy can be 100% with 100 Hz. But it not energy efficiency as shown in Figure 16(a). Therefore, we need make a trade off between energy consumption and recognition accuracy