Search

In this PhD thesis signal processing methods are developed for radar-based gait analysis. Analyzing the backscattered radio frequency waves from a moving non-rigid target in the time-frequency domain, reveals so-called micro-Doppler signatures. In the case of a walking person, these micro-Doppler signatures relate to gait kinematics by capturing velocity, acceleration, and rotation of individual body parts. Hence, they can be exploited for general gait classification and basic gait analysis. In the area of radar-based gait classification, a signal processing framework is developed to discriminate between five walking styles, including abnormal and cane-assisted gait. Toward this end, different joint-variable signal representations, including the spectrogram and the so-called cadence-velocity diagram, are adapted. These representations are used to identify physically interpretable features, and as input to automatic feature learning using principal component analysis. The thus obtained feature sets are evaluated and compared in terms of their corresponding classification performance. Additionally, a gait asymmetry detector is presented to identify differences between the left and right leg's motions from radar micro-Doppler signatures. The evaluation of the developed methods is based on realistic experimental radar data of 10 able-bodied individuals, and four persons with diagnosed gait disorders. Toward radar-based gait analysis, the thesis demonstrates the use of Doppler radar for measuring a set of medically relevant gait parameters, including stride time and maximal lower limb velocities. Here, a new method is presented for measuring the flight time through a radar device. In total, 11 biomechanical gait parameters are extracted and qualitatively and quantitatively assessed by using marker-based motion capture data. Further, parametric models for the radial velocities of the lower limbs during walking are developed. These models are then employed to estimate lowe

In this PhD thesis signal processing methods are developed for radar-based gait analysis. Analyzing the backscattered radio frequency waves from a moving non-rigid target in the time-frequency domain, reveals so-called micro-Doppler signatures. In the case of a walking person, these micro-Doppler signatures relate to gait kinematics by capturing velocity, acceleration, and rotation of individual body parts. Hence, they can be exploited for general gait classification and basic gait analysis. In the area of radar-based gait classification, a signal processing framework is developed to discriminate between five walking styles, including abnormal and cane-assisted gait. Toward this end, different joint-variable signal representations, including the spectrogram and the so-called cadence-velocity diagram, are adapted. These representations are used to identify physically interpretable features, and as input to automatic feature learning using principal component analysis. The thus obtained feature sets are evaluated and compared in terms of their corresponding classification performance. Additionally, a gait asymmetry detector is presented to identify differences between the left and right leg's motions from radar micro-Doppler signatures. The evaluation of the developed methods is based on realistic experimental radar data of 10 able-bodied individuals, and four persons with diagnosed gait disorders. Toward radar-based gait analysis, the thesis demonstrates the use of Doppler radar for measuring a set of medically relevant gait parameters, including stride time and maximal lower limb velocities. Here, a new method is presented for measuring the flight time through a radar device. In total, 11 biomechanical gait parameters are extracted and qualitatively and quantitatively assessed by using marker-based motion capture data. Further, parametric models for the radial velocities of the lower limbs during walking are developed. These models are then employed to estimate lowe

In this PhD thesis signal processing methods are developed for radar-based gait analysis. Analyzing the backscattered radio frequency waves from a moving non-rigid target in the time-frequency domain, reveals so-called micro-Doppler signatures. In the case of a walking person, these micro-Doppler signatures relate to gait kinematics by capturing velocity, acceleration, and rotation of individual body parts. Hence, they can be exploited for general gait classification and basic gait analysis. In the area of radar-based gait classification, a signal processing framework is developed to discriminate between five walking styles, including abnormal and cane-assisted gait. Toward this end, different joint-variable signal representations, including the spectrogram and the so-called cadence-velocity diagram, are adapted. These representations are used to identify physically interpretable features, and as input to automatic feature learning using principal component analysis. The thus obtained feature sets are evaluated and compared in terms of their corresponding classification performance. Additionally, a gait asymmetry detector is presented to identify differences between the left and right leg's motions from radar micro-Doppler signatures. The evaluation of the developed methods is based on realistic experimental radar data of 10 able-bodied individuals, and four persons with diagnosed gait disorders. Toward radar-based gait analysis, the thesis demonstrates the use of Doppler radar for measuring a set of medically relevant gait parameters, including stride time and maximal lower limb velocities. Here, a new method is presented for measuring the flight time through a radar device. In total, 11 biomechanical gait parameters are extracted and qualitatively and quantitatively assessed by using marker-based motion capture data. Further, parametric models for the radial velocities of the lower limbs during walking are developed. These models are then employed to estimate lowe

In this PhD thesis signal processing methods are developed for radar-based gait analysis. Analyzing the backscattered radio frequency waves from a moving non-rigid target in the time-frequency domain, reveals so-called micro-Doppler signatures. In the case of a walking person, these micro-Doppler signatures relate to gait kinematics by capturing velocity, acceleration, and rotation of individual body parts. Hence, they can be exploited for general gait classification and basic gait analysis. In the area of radar-based gait classification, a signal processing framework is developed to discriminate between five walking styles, including abnormal and cane-assisted gait. Toward this end, different joint-variable signal representations, including the spectrogram and the so-called cadence-velocity diagram, are adapted. These representations are used to identify physically interpretable features, and as input to automatic feature learning using principal component analysis. The thus obtained feature sets are evaluated and compared in terms of their corresponding classification performance. Additionally, a gait asymmetry detector is presented to identify differences between the left and right leg's motions from radar micro-Doppler signatures. The evaluation of the developed methods is based on realistic experimental radar data of 10 able-bodied individuals, and four persons with diagnosed gait disorders. Toward radar-based gait analysis, the thesis demonstrates the use of Doppler radar for measuring a set of medically relevant gait parameters, including stride time and maximal lower limb velocities. Here, a new method is presented for measuring the flight time through a radar device. In total, 11 biomechanical gait parameters are extracted and qualitatively and quantitatively assessed by using marker-based motion capture data. Further, parametric models for the radial velocities of the lower limbs during walking are developed. These models are then employed to estimate lowe