The Study Of Smartphone Sensor Based On Activity Recognition Algorithm

With the development of science and technology and computer technology, the wirelessnetwork is more and more widely popular. And wearable sensors, the product of newtechnology, are getting the recognition and popularity of scientific research personnel, thenthrough wireless sensor signal, developing human behavior recognition system has become aimportant research significance and been the valuable work. More and more researchinstitutions begin to use activity recognition system for a wide range of scientific research.More perfect functions of smart phones also bring great convenience to people’s daily life.Recent years, more and more attention is focused on the identification with sensors in thepeople’s activity, as the behavior recognition system in the field of health care demand,especially in the aged care, long-term health monitoring, and assist patients with cognitiveimpairment. Smartphones have a lot of softwares that can record the smartphone users dailyactivities. So this study puts forward a kind of Spectral clustering and Hidden Markov model(SC-HMM) based on daily activity recognition algorithm, which is on the basis of gainingthe sensor signals of smartphones. SC-HMM method use smartphones to collect sensor dataincluding the GPS location, acceleration information and the received signal strength, and iscombined with spectral clustering technology and the hidden markov model to study theacitivities, it can effectively and automatically recognize the user’s daily activities. This studymakes experiments on the real data to test the SC-HMM recognition method put forwardhere. The experimental results show that on the real smartphone data set, this method has highrecognition accuracy, and is superior to the previous traditional recognition methods. Thisstudy proposed identification method has a good practicability in the user behavior study,situational awareness, and other fields.The main contents of this research are as follows:(1) This research studies the architecture of wireless sensor network, and introduces thecharacteristics of the wireless sensor network and so on.(2) This research studies smartphone sensor technology, the acceleration sensor, RSSI,and further introduces the principle of RSSI, how to judge that the RSSI is abnormal and the reasons. And study how to collect the sensor data of the smartphone sensors and the analysisof the activities of smart phone users.(3) This research studies the classification of sensor based on activity recognition, whichhas two classes: sensor based on activity behavior recognition and sensor based onmulti-people activity recognition. The levels of recognizing the sensor based on activityrecognition are divided into there levels: at the lowerst level, collecting the sensor data, atintermediate level, adopting statistical inference, at the highest level, recognizing the goal orthe sub-goal of the activity. There are four main methods of sensor based on activityrecognition: probability reasoning method, logic and reasoning method, WiFi-basedrecogniztion method, the data mining-based method. This research mainly studies thesmartphone sensors based on activity recognition method.(4) This research studies the spectral clustering, which is clustering method in themachine learning. And the spectral clustering method is used to group the sensor datacollected from smartphones sensor into the similar K classes.(5) This research studies the unsupervised learning methods, features of hidden markovmodel and three problems it has to solve: estimation problem, decoding problem and learningproblems. The hidden markov model trains the activity time sequences to get smartphoneusers activities, and to recogize the activity of the untraining activity time sequences.Finally, make the conclusion of this research work, and look forward to the futureresearch work. Smartphone sensor based on SC-HMM recognition method put forward inthis study can be extended to recognize the interaction of two or more of the Smartphoneusers.