Unsupervised Learning : K-Means Approach In Classifying Human Walking Path

Shaidan, Muhammad Syahir (2016) Unsupervised Learning : K-Means Approach In Classifying Human Walking Path. Project Report. UTeM, Melaka,Malaysia. (Submitted)

Text (24 Pages) Unsupervised Learning K-Means Approach In Classifying Human Walking Path.pdf - Submitted Version Download (492kB)

Abstract

Along the decade, researchers have proposed some various methods to represents the human walking behaviors in either individual or in crowd environment.Human walking behaviors can be divided into several viewpoints such as route selection,navigation,and path finding.Although human walking behavior is unpredictable and has dynamic characteristics,these viewpoints are the most dominant behavior that human always considered when they are walking.Furthermore,many researchers have widen their scope of investigation as they are more focusing in human walking behaviors under panic situations that usually in crowded conditions due to natural disaster or terrorist attacks.In this project,a method to analyze human walking data is proposed and human walking data will be analysed by classifying the path taken by human while walking whether right,left of a forward path.However in order to collect human walking data,an experiment is set up which involves five different subjects with different gender,height and weight considered.In the experiment,the subject will wear a wearable device containing inertial measurement unit around their waist and walks along a created path.Inertial measurement unit is used to determine the orientation and position of human while they are walking.The data considered from inertial measurement unit is Yaw,Pitch,and Roll data.The human walking data will be classified by using unsupervised learning method because human walking data are unclassified data and the results of the analysis are cannot be predicted.K-means clustering method is used to classified human walking activity.The number of cluster will be determined by using K-means clustering.After that, the performance measurement of K-means clustering is carried out to evaluate the performance of K-means clustering.Silhouette Coefficient method is used for this purpose.The validity of number of cluster for clustering human walking activity is determined.As conclusion,the number of cluster that is suitable to classified human walking data is three.

Actions (login required)

View Item