Classification of elderly subjects to fallers and non-fallers using a novel set of features extracted from gait frieze pattern

Changes in gait pattern are early symptoms in many disorders such as balance and control problems resulted in fall among elderlies. This paper aims at proposing a new set of features extracted from Gait Frieze Pattern (GFP) in order to classify seniors to fallers and non-fallers. For indicating the effectiveness of the presented method, the algorithm is used for recognition of different type of abnormal gaits. The introduced method consists of three main steps: extracting the subject from background, generating GFP and aligning them, and building the proposed image from GFP by thresholding followed by morphological operations. For evaluation of the proposed features, video sequences are collected from 8 elderly fallers, 8 non-fallers, and 8 youth while performing standard Timed Up and Go (TUG) test. In addition to TUG test youths are asked to walk fast and pretend to walk with 6 different types of abnormalities (limping, waddling, anterior- posterior sway, lateral sway, dragging, steppage gait). For finding correct classification rate, each time one data is considered as test and others as train and label of train data with the most similarity with test one on the score of normalized cross correlation is assigned to test data. Comparing to conventional TUG test, correct classification data is improved around 20% for faller detection. In addition, correct classification rate for detecting of different abnormalities in gait is approximately 90%.