Non-Intrusive Gait Recognition Employing Ultra Wideband Signal Detection

A self-regulating and non-contact impulse radio ultra wideband (IR-UWB) based 3D human gait analysis prototype has been modeled and developed with the help of supervised machine learning (SML) for this application for the first time. The work intends to provide a rewarding assistive biomedical application which would help doctors and clinicians monitor human gait trait and abnormalities with less human intervention in the fields of physiological examinations, physiotherapy, home assistance, rehabilitation
success determination and health diagnostics, etc.
The research comprises IR-UWB data gathered from a number of male and female participants in both anechoic chamber and multi-path environments. In total twenty four individuals have been recruited, where twenty individuals were said to have normal gait and four persons complained of knee pain that resulted in compensated spastic walking patterns. A 3D postural model of human movements has been created from the backscattering property of the radar pulses employing understanding of spherical trigonometry and vector fields. This subjective data (height of the body areas from the ground) of an individual have been recorded and implemented to extract the gait trait from associated biomechanical activity and differentiates the lower limb movement patterns from other body areas.
Initially, a 2D postural model of human gait is presented from IR-UWB sensing phenomena employing spherical co-ordinate and trigonometry where only two dimensions such as, distance from radar and height of reflection have been determined. There are five pivotal gait parameters; step frequency, cadence, step length, walking speed, total covered distance, and body orientation which have all been measured employing radar principles and short term Fourier transformation (STFT). Subsequently, the proposed gait identification and parameter characterization has been analysed, tested and validated against popularly accepted smartphone applications with resulting variations of less than 5%. Subsequently, the spherical trigonometric model has been elevated to a 3D postural model where the prototype can determine width of motion, distance from radar, and height of reflection. Vector algebra has been incorporated with this 3D model to measure knee angles and hip angles from the extension and flexion of lower limbs to understand the gait behavior throughout the entire range of bipedal locomotion. Simultaneously, the Microsoft Kinect Xbox One has been employed during the experiment to assist in the validation process. The same vector mathematics have been implemented to the skeleton data obtained from Kinect to determine both the hip and knee angles.
The outcomes have been compared by statistical graphical approach Bland and Altman (B&A) analysis.
Further, the changes of knee angles obtained from the normal gaits have been used to train popular SMLs such as, k-nearest neighbour (kNN) and support vector machines (SVM). The trained model has subsequently been tested with the new data (knee angles extracted from both normal and abnormal gait) to assess the prediction ability of gait abnormality recognition. The outcomes have been validated through standard and wellknown statistical performance metrics with promising results found. The outcomes prove the acceptability of the proposed non-contact IR-UWB gait recognition to detect gait.