Do you need this or any other assignment done for you from scratch?
We have qualified writers to help you.
We assure you a quality paper that is 100% free from plagiarism and AI.
You can choose either format of your choice ( Apa, Mla, Havard, Chicago, or any other)
NB: We do not resell your papers. Upon ordering, we do an original paper exclusively for you.
NB: All your data is kept safe from the public.
Introduction
Gait recognition has gained importance in biometric related authentication systems used in wearable tech. Gait recognition is unobtrusive, and helps to capture important health data of users. The assumption is that a person who is physically fit will have a certain type of unique gait, captured through signature points, and this gait will change when the person falls sick or if he is physically disabled. Healthcare systems wearable tech firms have introduced implantable Medical Devices (IMD) with gait recognition systems having built in accelerometers, which are fitted on body sensors connected to the wearable tech. The argument posed in the paper is that battery life for smart phones and other wearable tech is an issue since the batteries have to be recharged externally.
However, IMD implies that medical devices are implanted inside the body and surgery is needed to remove the batteries. While these systems provide a reliable method to monitor gait and related health parameters, the main problem is the large energy consumed in continuous sampling of signals. The article examined in this paper proposes a Kinetic Energy Harvester (KEH) Gait that can reduce power consumption of batteries by 78.15%. The authors present findings from studies where two types of KEH hardware were used on 20 subjects to study the output voltages from the devices. Two prototypes were constructed, based on piezoelectric energy harvester (PEH) and electromagnetic energy harvester (EEH). The authors report that KEH Gait devices used in the experiment showed 10-25-% lower accuracy in authentication, while reduction in power use was 78.15%. To overcome this problem of lack of accuracy, the authors propose a Multi-Step Sparse Representation Classification (MSSRC), which integrates data from multiple steps and gives the desired level of accuracy, while consuming less power (Xu et. al 1).
Novel Idea
The novel idea that the authors present in this paper is to use energy harvester devices to generate power when users move their arms, generating strain that is converted into electrical voltage. This voltage is used to power the wearable tech device, used for health monitoring. This idea is interesting since it implies that batteries on wearable tech can be eliminated or smaller batteries can be used. In addition, frequency of recharging is reduced (Xu et. al 1).
Motivation
The authors are motivated by finding a solution to reduce power consumption of wearable tech devices, used in healthcare applications. The problem was to use KEH as a energy harvesting device as well as a gait authenticating system.
Approach and Methodology
The authors use a combination of literature review and experiments to present their findings. First, the authors introduce the subject of gait recognition in wearable tech, next they provide a technical background on devices used for energy harvesting. The authors then speak of trust and attacker models of authentication system, and then present the system architecture of the model used in the research. The authors develop prototypes of KEG Gait and present data collected through experiments (Xu et. al 2). A detailed explanation of the methods and processes used are described a follows.
The KEH system has three parts: offline dictionary training module, preprocessor of inputs, and classification. The first time it is used, gait cycles are separated from voltage and included in the same length, and deleted cycles are categorized as unusual cycles. The accepted set of gait cycle signals is used to create the training dictionary. A projection optimization algorithm is used to developed an optimized projection value Ropt and the new training dictionary A~ = RoptA is used as the classifier. The gait cycle is applied for test signal acquisition to derive the gait cycles from the test signals. To recognize the gait signal, the time series of subjects walking is split into different segments so that each segment carries the full gait cycle. Since the gait cycle is a stride made of two steps, the two steps are combined. After the gait cycle is extracted, voltage data is split into short gait cycles using peak detection for cycle distribution. Linear interpolation is used to normalize cycles into equal length since equal length vectors are needed (5). Multi-Step Sparse Representation Classification (MSSRC) is used to remove the accuracy variation of results, where PEH gives 86% accuracy and EEH give 75% efficiency. MSSRC links the sparse coefficient vector obtained from different gait cycles to improve accuracy of recognition (Xu et. al 6). The next section presents data analysis and performance evaluation from the research.
Performance Evaluation
The authors have presented and evaluated performance and data gathered by the system.
Validation using experiments
The author used a series of experiments with volunteers to test the KEH device. Three types of vibration based electric energy conversion devices were examined; these are piezoelectric, electromagnetic, and electrostatic. The last option of electrostatic based devices was not considered since it needs an external power source. Two energy harvesters with PEH and EEH were developed. PEH device is used to convert mechanical strain into electric voltage, and the strain is obtained from body movement of the user. In the devices, an extended arm, coated with a piezo electric coat, resonates when vibration occurs at the root, generating an AC voltage. In the EEH device, voltage is developed when an electric conductor passes through a magnetic field, induced by a light weight magnet. This voltage is fed to the IMD (Xu et. al 2).
Prototype for the PEH data logger was built by using Volture, a vibration energy harvesting item, manufactured by MIDE Technology. This component has a transducer that generates AC voltage. A 2-axis accelerometer was used to note the acceleration signals along with the voltage. Another device, Arduino Uno, is powered by a 9 volts battery, was used as the microcontroller to sample data. Data gathered was stored on an 8GB micro SD card. An on/off switch and another switch to start and stop data collection was used to control data gathering. Arduino measures voltage in the 0-6 volts at 10 bits resolution, providing 1024 discrete values. Voltage was measured with the formula V = (5x volts)/ 1023. EEH data logger gathered voltage obtained from the EEH device. Microcontroller was built from the Tmote sky board to gather inductor generated data at a sampling rate of 100Hz. Data gathered was saved to a 48K Flash drive of the microcontroller and a pair of AA batteries were used to provide power for the sky board (Xu et. al 7).
Data collection methods are described as follows. For the research, 20 people, 14 males and 6 females, were selected. Each person was tested through two sessions, with one week intervals. Data logger devices were given to them and they were asked to hold it and walk at a normal speed of about 1 meter/ second. The process was carried out at in-indoor and outdoor location to assess the factor of external conditions. The location selected varied from plain ground, asphalt, and grass surfaces. Subjects were asked to walk for five minutes. This variation in terrain helped to capture differences in natural gait over an interval of one week under different ground conditions. For each subject, 300 seconds of data was collected from the EH and accelerometers. Two types of voltage datasets were obtained by using the EEH and PEH instruments. During the tests, gait cycle segmentation was recorded while unusual gait cycle was deleted to obtain 200 gait cycles for each person (Xu et. al 7)
Three important metrics were examined. Recognition accuracy is the ratio of numbers of true classification to total tests. False positive rate (FPR) is the number of access requests accepted that are false. False negative rate (FNR) is the number of rejected access requests that are genuine. Recognition accuracy was calculated as the output voltage in one cycle of MSSRC using Support Vector Machine (SVM), K-Nearest Neighbor (KNN), and Naive Bayes (NB). Results show that MSRC has better results than SVM and KNN, for tasks such as face and voice recognition (Xu et. al 8).
The relation between recognition accuracy and power consumption was measured since power consumption and sampling rate are directly related. MSSRC was used as a classifier to calculate recognition accuracy for multiple sampling rates from 100-1Hz. It was seen that the accuracy of recognition rose with increased sampling rate. This indicates that the learning dictionary became more powerful when sampling rate increased. The optimum sampling rate was at 40Hz, after which the efficiency decreased. The accelerometer system consumed 300 micro watts at 40 Hz, an amount beyond the capacity of PEE and EEH. Studies show that EEH generated 10.17 micro watts during walking and PEH generated 1 micro watt from human activity harvesting. This in turn implies that the efficient sampling rate of 40Hz will not be reached and recognition accuracy decreased. (Xu et. al 9).
For the test, the researchers set the value of k at 1, changed the compression rate from 15% to 100% to calculate the recognition accuracy. It was seen that the recognition accuracy at 8Hz from the voltage signals was higher than that obtained from the accelerometer. MSSRC was used to improve the recognition accuracy of harvested power signal and the level of recognition was acceptable to the reading for raw accelerometer. Compression ratio was set at 75% and KEH Gait recognition accuracy was measured when voltage was increased from 1 to 8. Tests were carried out for EEH and PEH and it is seen that EEH is better by 6% when compression rate is 75% and 4% better when K is 5. The results indicate that performance of EEH is less than PEH (Xu et. al 10). Please refer to the following figure that illustrates test results.
An important test conducted was to evaluate energy consumption. The focus was to find ways to reduce energy consumption for three areas of sensor sampling, memory read/ write, and transmission of data. Among these three operations, least energy consumed was from read/ write. Results show that to collect 5 seconds of data at 40Hz, only 40.6nJ is needed, therefore, and energy consumed by SRAM is very small. Sensor power sampling was done in two instances and these are for sampling accelerometers and sampling KEH. Tests were carried out for different states from S1 to sleep. Results indicate that for a sampling rate of 40Hz, KEG Gait consumes 17.38 micro watts while accelerometer based system consumed 230.74 micro watts. However, 3 gait cycles are needed for accelerometer systems, consuming 86.9mJ and 5 gait cycles for KEH systems, consuming 696.22 mJ, indicating that KEH consume more power (12). Energy consumption for data transmission shows that KEH creates 200 voltage samples in five seconds, giving 300 data bytes and the total energy consumed is 106.08 mJ. The accelerometer collets data every 3 seconds, giving 540 bytes and consuming 190.94mJ. An analysis of total power consumed presents a different picture. Results show that KEH consumes 192.98 mJ while accelerometers system uses 993.16mJ, thus making KEH more energy efficient (Xu et. al 13). The next section presents information about related work.
Related Work
The authors report that gait recognition is done as per three methods, vision based, floor sensor based, and wearable sensor based. Vision based systems use video cameras to record and recognize gait. Gait frames are extracted and used to define the gait. Floor sensor system use force plates, installed under the floor and used to capture gait by measuring force of ground reaction. Acceleration and wearable sensor methods are more accurate, they measure gait directly, and they cannot be spoofed easily and they are not subject to view point changes.
The authors have used a novel way to overcome the problem of power consumed by wearable tech devices and authentication system. The use of KEH harvester is in use since many years. However, this is the first time that KEH principle is applied to gait authentication system. If the results are widely accepted, then devices in the future would have smaller batteries, the frequency of recharging would be extended, and the devices would be able to self authenticate with higher accuracy.
Conclusions
KEH presents acceptable level of accuracy for user authentication and power consumed is significantly less. Of important is the question of using PEH or EEH. A detailed analysis of the article shows that PEH systems typically weigh 23.5 grams, they are thicker than EEH, consume 1 micro watt, give an accuracy of 86.1% and cost about 157 USD. EEH on the other hand, weigh 65 grams, give an accuracy of 75.2%, consume 19.17 micro watt, and cost 37.5 USD. Therefore, PEH has a higher recognition and authentication ability. Problem of higher weight of EEH is due to magnets used in the devices.
Strengths and weaknesses
The weakness of the research is that a major part deals with authentication and verification mechanisms, to prevent spoof attacks. While prevention of fraud is important, the objective of the paper was to explore methods used to reduce battery power consumption, and this aspect is not the focus of the research. In any case, the advantages of KEH Gait over traditional accelerometer based systems are clear. Some anomalies are evident in the research method and the manner in which the key theme of energy conservation is used. Gait formation is due to a number of factors, not related to the health. These can include type of shoes worn, speed of walking, time when the tests are conducted since during early morning walks, a person is not tired and takes longer strides, while in the evening, the strides are smaller and less forceful.
These variations were not considered in the tests. Another aspect is that vibrations from heavy traffic, from bus and trains during commuting can be significant. They can power the devices, charge other components, and give false positive results. Data interpretation and gait recognition would show that the user is an imposter and data servers would block the user. The size and weight of PEH and EEH appear quite bulky. PEH system used in the tests weigh 23.5 grams and has a size of 4.6 x 3.3 x 0.1 cms while EEH weigh 65 grams and have a size of 4.7 x 0.8 x 0.8 cms. These devices were carried by the respondents in their hands, since hand movement is maximum and produces the required level of mechanical strain, required to power the devices. It remains to be seen if KEH would generate the required level of power if they are mounted on the back or placed in the pocket. There is also a possibility that EEH can be impacted by magnetic sources such as mobile phones, electric power lines, microwaves, and other devices, and these can create false positives.
The strength is that EEH and PEH systems offer is in energy and power generation. This development is of interest since power consumption of accelerometers need 3 gait cycles for sampling while KEH required 5 cycles. The min energy saving appears to come from energy consumption of data transmission, where accelerometers consume 190.94mJ since they sample and send data in 3 seconds while KEH consumes 106.08mJ since it samples every 5 seconds. Given these observations, it would make more sense to configure the accelerometer to sample every 5 seconds, thus consuming less power. The authors claim that accuracy of sampling rate increases at 40Hz compression rate. However, there is no control in the device that caps this rate.
Improvement suggestions
Some suggestions for improvement are that the sleep mode and ‘wake up’ mode of the devices should be increased so that they start transmitting data faster. A longer sleep mode will mean that the user, who can be a patient suffering from heart attacks, would be suffering from an attack, and the staggered gait he would take when under an attack, would not be recorded in the learning dictionary. These would be regarded as false positive and the user would be locked out of the system, at a time when the IMD is needed the most. With an accuracy of 86.1% for PEH and 75.2% for EEH, it is clear that the authentication system is not reliable. The implication is that these systems cannot resist passive imposter attacks and active imposter attacks. Accelerometer based systems offer accuracy of 95% or above. The suggestion is to have a small standby supply of a few mini watts that will keep the system on standby.
The research has opened up a new area for further development for wearable tech, and this is to generate power through vibrations and movement of the body. This is especially true for IMD devices that cannot be charged easily since they are embedded in the body. KEH promises a new area for development and research, and methods to conserve battery life using EEH and KEH devices indicate the way forward for technology. While KEH is a very good concept, it is essential to find ways to improve the accuracy at all compression and sampling rates. Main suggestions are to improve the reaction rate, reduce sampling rate, revisit data transmitted and only send important data, and reconfigure KEH to generate more power so that batteries are eliminated.
Work Cited
Xu, Weitao , et. al. “KEH-Gait: Towards a Mobile Healthcare User Authentication System by Kinetic Energy Harvesting”. NDSS’17, University of New South Wales, Australia, 2017, pp. 1-15
Do you need this or any other assignment done for you from scratch?
We have qualified writers to help you.
We assure you a quality paper that is 100% free from plagiarism and AI.
You can choose either format of your choice ( Apa, Mla, Havard, Chicago, or any other)
NB: We do not resell your papers. Upon ordering, we do an original paper exclusively for you.
NB: All your data is kept safe from the public.