Activity Recognition with Smartphone Tri-axial Accelerometer

• Main Authors: Chin-Ting Chu, 朱晉廷
• Other Authors: 許輝煌
• Format: Others
• Language: zh-TW
• Published: 2015
• Online Access: http://ndltd.ncl.edu.tw/handle/19963605327719675547

碩士 === 淡江大學 === 資訊工程學系碩士班 === 103 === Human activity recognition is an important research topic, there are many solution such as using wearable sensor to human activity recognition, but also using image processing to human activity recognition. Image processing solution is in a fixed location to human activity recognition, does not apply to personal case. The solution of using wearable sensor to human activity recognition is inconvenient in action, therefore not suitable for use in long time detection. Along the development of smartphone, and its convenience. Currently it can be said that everyone has a smart phone, and the smartphone is equipped with a variety of sensors, and processing operations ability. This makes the smartphones very suitable replace wearable sensors for long time activity recognition, related research and application is also increasing. However, in these research, many of them have some restrictions, for example, the smartphone needs to be fixed at some location, like pants pocket or jacket pocket ... etc. or needs to be fixed direction. These restrictions apply in daily life will be very inconvenient, therefore, in this paper, we propose an activity recognition system based on phone’s position recognition method, first we use the gyroscope to distinguish position, according to the position use different classification models, this step can solve the problem of restricted smartphone position, and then using tri-axial accelerometer inside the phone to get the acceleration value, and calculates the sum of squares of the tri-axis acceleration value to obtain the whole acceleration, these new feature can solve the problem of restricted smartphone orientation, finally, we use support vector machine to activity recognition, the active are walk, run, up-stair, down-stair, stand/sit, and the result and the time displays on the screen. The experimental results show that we used phone’s position recognition method can upgrade accuracy, and we can ignore the impact of phone’s orientation. Therefore, we believe that using our proposed activity recognition system can solve the problem of the restricted smartphone location and restricted smartphone direction.