根據慣性測量單元的走路運動辨識

IMU-Based Walking Workouts Recognition

陳國健(Wahjudi, Fanuel)

指導教授 ： 林甫俊

國立交通大學/電機學院/電機資訊國際學位學程/碩士(2018年)

摘要

近來，隨著物聯網（IoT）的出現，穿戴式設備的發展正在快速增長。 這些穿戴式設備與使用者互動，且通常與慣性感測器整合，包括加速度計和陀螺儀。此研究的重點在於利用固定於腳上的慣性感測器識別步行運動，因為它可在醫療保健領域中有其他實用的應用，例如準確估計鍛煉期間所燃燒的卡路里。
我們的目標是根據從固定於腳上的IMU所收集的數據，以識別步行運動，包括在不同道路環境的行走和快走，例如平地、樓梯或者沒有樓梯的斜坡。此外，如果是斜坡或樓梯，還需要區分移動方向（向上或向下）。所有這些都是維持人們健康的普遍運動，每種都耗費不同程度的力氣，因而導致不同程度的卡路里燃燒。
我們專注於（1）提出一種用於識別擴展平足(EFF)階段的演算法，可進一步用於生成機器學習演算法所需的特徵值的界線，以及（2）選擇用於檢測特定類型的步行運動的典型特徵。所應用的機器學習模型包括決策樹（Decision Tree）、隨機森林（Random Forest）和K-近鄰演算法（K-Nearest Neighbor）。我們將評估這些模型，並確定哪一個模型最符合我們的研究目標。據我們所知，並沒有其他步伐分析的研究，能夠如此精細地識別這些步行運動。

關鍵字

活動識別 ； 環境檢測

並列摘要

Recently, with the emergence of the Internet of Things (IoT), the development of wearable devices is growing fast. These wearable devices are used to interact with the user and often integrated with inertial sensors including accelerometer and gyroscope. This research focuses on recognizing walking workouts based on the foot-mounted inertial sensor as it can provide useful applications in healthcare areas such as accurate estimate of calories burnt during exercise effort.
Our objective is to recognize walking workouts including walking and brisk-walking under different track environments such as flat surface, a staircase or a slope without stair based on the data collected from a foot-mounted IMU. Moreover, if it is a slope or a staircase, the direction of movement (upward or downward) need also to be distinguished. All these are popular workout activities in maintaining people’s health and each of them requires a different level of effort which leads to a different level of calories burnt.
We focus on (1) proposing an algorithm for identifying the extended foot-flat phase which can be further used as a boundary to generate features required by machine learning algorithms, and (2) selecting typical features for detecting a particular type of walking workout activity. The machine learning models to be applied include Decision Tree, Random Forest and K-Nearest Neighbor. We will evaluate these models and determine which one works the best for our research objective. To our best knowledge, no gait analysis studies have been done to recognize these walking workouts with such fine granularity

並列關鍵字

Gait Analysis ； Activity Recognition ； Environment Detection ； IMU Sensor

參考文獻

[1] W. Tao, T. Liu, R. Zheng, and H. Feng, “Gait analysis using wearable sensors” in Sensors 2012, 12, pp. 2255–2283, February 2012.

[2] V. N. T. Sang, T. Q. D. Khoa, S. Yano, and T. Kondo, “Ascending and Descending Walking Recognition using Smartphone” in MOBIHEALTH'15 Proceedings of the 5th EAI International Conference on Wireless Mobile Communication and Healthcare, pp. 120-123, London, Great Britain, October 2015.

[3] S. Sprager, and M. B. Juric, “Inertial Sensor-Based Gait Recognition: A Review.” In Sensors 2015,15, pp. 22089-22127, September 2015.

[4] N. Ravi, N. Dandekar, P. Mysore, and M.L. Littman, “Activity Recognition from Accelerometer Data” in the 17th conference on Innovative applications of artificial intelligence, Volume 3, pp. 1541-1546, Pittsburgh, Pennsylvania, July 2005

[5] M. Ghobadi and E. T. Esfahani, “Foot-Mounted Inertial Measurement Unit for Activity Classification” in 36th Annual International Conference of the IEEE Engineering in Medicine and Biology Society, pp. 6294-6297, Chicago, 2014.