【摘要】：隨著物聯(lián)網(wǎng)時代的到來,基于傳感器的活動識別研究成為熱點,其中利用可穿戴式傳感器的活動識別在移動計算領(lǐng)域的研究較多,而基于非穿戴式傳感器的活動識別研究更適合于智能環(huán)境應(yīng)用中,這也是進(jìn)一步完善智能家居設(shè)計乃至實現(xiàn)的關(guān)鍵點。而且,考慮到使用方便性和設(shè)備的無打擾性,基于非穿戴式傳感器的活動識別也具有其自身的優(yōu)勢,到目前為止,基于非穿戴式傳感器的活動識別仍然具有很大的挑戰(zhàn)性。因此,本文研究基于非穿戴式傳感器的多人室內(nèi)活動識別,其中,非穿戴式傳感器包括環(huán)境傳感器和物體傳感器,室內(nèi)活動主要包括一些常見的日常活動,這些活動可能會對環(huán)境或是相應(yīng)物體產(chǎn)生影響,包括:在工位學(xué)習(xí)、打印、燒水、煮咖啡、談話、讀書、寫字、用電腦、喝水。故本文的研究根據(jù)活動產(chǎn)生影響的不同可分為兩個部分,基于環(huán)境傳感器的活動識別研究和基于物體傳感器的活動識別研究。在基于環(huán)境傳感器的活動識別研究,本文利用環(huán)境傳感器,包括電表、溫、濕度傳感器、紅外人體感應(yīng)傳感器、避礙傳感器、光強傳感器和錄音筆搭建模擬實驗平臺采集活動數(shù)據(jù),利用采集到的數(shù)據(jù),本文針對數(shù)據(jù)形態(tài)的不同及多人并發(fā)活動的特點,給出兩種算法和模型進(jìn)行活動識別,分別為基于動態(tài)時間扭曲(DTW)的K近鄰(KNN)模型和多標(biāo)簽算法模型的基于支持向量機(jī)(SVM)的多分類器模型,最后證明,多標(biāo)簽算法模型的基于SVM的多分類器模型表現(xiàn)更好,最終得到了89%的識別準(zhǔn)確率。在基于物體傳感器的活動識別研究中,本文主要應(yīng)用的傳感器設(shè)備為射頻識別(RFID)標(biāo)簽和RFID讀寫器,其中標(biāo)簽有兩種,貼附于各個物體表面,RFID讀寫器用來讀取各個標(biāo)簽的接收信號強度(RSS)數(shù)據(jù)。本文設(shè)計模擬平臺采集RSS數(shù)據(jù)。利用采集到的活動數(shù)據(jù),在此部分,考慮到RSS數(shù)據(jù)的特點,本文提出數(shù)據(jù)映射方法,再利用CNN-LSTM模型對數(shù)據(jù)進(jìn)行自動的數(shù)據(jù)空間特征和時間特征提取,并最終在活動識別上達(dá)到了95%的識別準(zhǔn)確率。在本文的最后,分別對基于環(huán)境傳感器和物體傳感器的活動識別方法做了完整的測試和對比分析,從模擬環(huán)境搭建策略到數(shù)據(jù)采集過程,再到活動識別算法模型,對基于非穿戴式傳感器的活動識別給出總體的識別和設(shè)計方案。綜上所述,本文給出基于非穿戴式傳感器的多人室內(nèi)活動識別方法,方法產(chǎn)生了可接受并優(yōu)于以往相似研究的識別結(jié)果,并具有很好的可擴(kuò)展性和可移植性。
[Abstract]:With the advent of the Internet of things era, the research of sensor-based activity recognition has become a hot topic, in which wearable sensors are widely used in the field of mobile computing. The research of activity recognition based on non-wearable sensors is more suitable for intelligent environment application, which is also the key point to further improve the design and implementation of smart home. Moreover, considering the convenience of use and the non-disturbance of the device, the activity recognition based on the non-wearable sensor has its own advantages. So far, the activity recognition based on the non-wearable sensor is still very challenging. Therefore, this paper studies the identification of multi-person indoor activities based on non-wearable sensors. Among them, non-wearable sensors include environmental sensors and object sensors, and indoor activities mainly include some common daily activities. These activities may have an impact on the environment or related objects, including: studying, printing, boiling water, making coffee, talking, reading, writing, using computers, drinking water. Therefore, the research of this paper can be divided into two parts according to the different influence of activity, namely, the research of activity recognition based on environmental sensor and the study of activity recognition based on object sensor. In the research of activity recognition based on environmental sensors, this paper uses environmental sensors, including ammeter, temperature, humidity sensor, infrared sensor to avoid obstacles. Light intensity sensor and recording pen are used to build a simulation experiment platform to collect activity data. According to the different data forms and the characteristics of multiple concurrent activities, two algorithms and models are presented to identify the activity. It is a K-nearest neighbor (KNN) model based on dynamic time-distorted (DTW) and a multi-classifier model based on support vector machine (SVM) algorithm model. Finally, it is proved that, The multi-classifier model based on SVM is better than the multi-label algorithm model, and the recognition accuracy is 89%. In the research of object sensor based activity identification, the main sensor devices used in this paper are RFID (RFID) tag and RFID reader. There are two kinds of tags attached to the surface of each object. The RFID reader is used to read the received signal strength (RSS) data for each tag. This paper designs a simulation platform to collect RSS data. In this part, considering the characteristics of RSS data, this paper proposes a method of data mapping, and then uses the CNN-LSTM model to extract automatically the spatial and temporal features of the data. Finally, 95% recognition accuracy is achieved in activity recognition. At the end of this paper, the methods of activity recognition based on environment sensor and object sensor are tested and compared, from the strategy of simulation environment to the process of data acquisition, and then to the algorithm model of activity recognition. The overall recognition and design scheme of the motion recognition based on non-wearable sensor is given. To sum up, this paper presents a multi-person indoor activity recognition method based on non-wearable sensors, which produces acceptable and better recognition results than previous similar studies, and has good scalability and portability.
【學(xué)位授予單位】：哈爾濱工業(yè)大學(xué)
【學(xué)位級別】：碩士
【學(xué)位授予年份】：2017
【分類號】：TP212.9

【相似文獻(xiàn)】

相關(guān)期刊論文 前10條

1 ;穿戴式電腦[J];應(yīng)用科技;2000年07期

2 ;可穿戴式計算機(jī):一切都有可能[J];中國教育網(wǎng)絡(luò);2012年12期

3 ;美推出可穿戴防彈PC[J];軍民兩用技術(shù)與產(chǎn)品;2002年07期

4 劉金劍;謝慶森;;3C產(chǎn)品的穿戴式設(shè)計初探[J];藝術(shù)與設(shè)計(理論);2010年09期

5 范曉東;;穿戴式設(shè)備興起:科技想要擁抱生命?[J];互聯(lián)網(wǎng)周刊;2013年05期

6 沈建苗;;認(rèn)清可穿戴式計算的真正本質(zhì)[J];微電腦世界;2013年11期

7 王啟明;;一場穿戴式技術(shù)革命正在北美地區(qū)興起[J];全球科技經(jīng)濟(jì)w，

本文編號：2389391