Human gait identification using two dimensional multi-resolution analysis

Abstract

Application of wearable devices like as inertial sensors has grown rapidly in various daily living applications such as, in wearable gait analysis due to several factors, such as being easy-to-use and low-cost. Many methods have been suggested to extract various heuristic and high-level features of motion sensor data to identify discriminative gait signatures to distinguish the target individual from others. In this study, we propose an identification method for human gait identification using spectro-temporal two dimensional (2D) expansion of human gait cycles, and then implement a 2D multi-resolution analysis to decompose the expanded 2D space to different levels of resolution. We propose a systematic methodology for processing motion signals for the purpose of human gait identification by first extracting the Gait cycles. Then, we perform 2D spectro-temporal representation of each cycle. After that, we conduct multi-resolution decomposition and analysis of the 2D spectrograms to extract useful information for the predictive task. We collect raw motion data from five inertial sensors placed on the chest, lower back, right hand wrist, right knee and right ankle. After pre-processing the raw recordings, we propose an effective heuristic segmentation method to extract the gait cycles from the processed data. Spectro-temporal features are extracted by merging key instantaneous spectral descriptors in a gait cycle that characterize the non-stationarities in each gait cycle inertial data in two dimensions. The 2D spectro-temporal expansion of the gait cycles extracted from inertial sensor data from a population of 10 subjects are decomposed to several levels of resolutions. Based on the extracted features from each level of decomposed space, the identification task is accomplished. Based on our experimental results, 93.36% subject identification accuracy was achieved.