Forward scattering radar for real-time detection of human activities and fall classification

Falls pose a considerable threat that raises concerns among the elderly aged 65 and above worldwide. Many approaches have been used such as wearable and non- wearable sensors to observe their activities for mitigation of falling event consequences. However, these approaches have many constraints especially those related to false alarms, light sensitivity, occlusion, and personal privacy invasion. Radar technologies work independently of light, weather, and occlusion. They also provide advantages of a very low false alarm rate, and preserving personal privacy. The purpose of using radar is to detect, identify and recognize the moving targets based on the Doppler phenomenon generated from the interaction of the transmitting electromagnetic waves with the moving parts of the target. Two main types of radar technologies are available, the monostatic and bistatic radars. In bistatic radar, the bistatic angle plays a major role in measuring radar Doppler frequency. Forward scattering radar (FSR) is described as a special case of bistatic radar when the bistatic angle is near to 180°. The FSR system has some outstanding features compared to the conventional monostatic and bistatic radar including, enhancing radar cross-section of the target, relatively simple hardware, high ability to detect low observable targets regardless of the target’s shape and coating materials, and the absence of signal phase fluctuations. Accordingly, the aim of this thesis is to use the FSR for detecting and differentiating fall events among other human daily activities. The aim is divided into four objectives that were pursued in this thesis. The first objective is a mathematical characterization of the translational, vibrational, and rotational motions that describe the movements of human body parts. The second objective is to model and simulate the movement of human arms, legs, and torso. The third objective is the experimental analysis of the detection and identification of human daily activities. Finally, the classification of fall events among these activities based on measuring the radar Doppler and micro-Doppler signatures. Five activities, namely walking, sitting on a chair, squatting, standing from a chair, and bending, and five types of falls from different postures, namely, forward falling, backward falling, falling while trying to sit on a chair, falling while trying to stand up from a chair, and falling from walking, were experimentally conducted indoor to extract and analyze the human Doppler and micro-Doppler signatures. The frequency spectrum signatures extracted from the experimentally collected data were used as input features to the classification unit. Support Vector Machine (SVM) algorithm modified by kernel linear function was used for classifying the fall event from the other activities. Furthermore, the prediction ability of the classification process was improved by measuring the average value of the unseen example. The experimental results revealed that the FSR system has the ability to detect and differentiate the low-speed human body motions when performing daily activities and fall events. Remarkable classification accuracy of 99.98% was obtained for classifying of different types of fall events from other daily activities. Finally, the FSR system supported by the SVM classifier can be considered as a promising assistive-living sensor for real-time monitoring of human movements.

Text FK 2020 6 - ir.pdf Download (1MB)

Actions (login required)

View Item