Reference #: 1756
The University of South Carolina is offering licensing opportunities for A Contactless and Camera-Free Technology for Gait Assessment.
Background:
Human gait, the pattern of walking, involves the movement of the body’s center of gravity and serves as an essential indicator of overall health. Metrics such as gait speed, step length, stride length, and cadence, are critical for evaluating patient health, with gait speed often termed the ”sixth vital sign.” Traditional methods of gait analysis, like observing patients during clinical visits or using motion capture systems (MOCAPS), offer high accuracy but are costly, time-consuming, and confined to specific setups, making them impractical for home use. Vision-based systems, though cost-effective, perform poorly in low-light conditions, while wearable sensors can be uncomfortable and inconvenient, particularly for elderly patients. Innovations like WiTrack and WiGait use radio frequency (RF) signals for contactless monitoring, but their performance is limited by the low resolution of Wi-Fi signals, especially for patients with abnormal gait patterns, such as those caused by stroke. Millimeter wave (mmWave) technology addresses many limitations of earlier systems by offering high-resolution tracking, even in complete darkness, with a compact and cost-effective setup. Unlike vision-based systems, mmWave technology protects privacy and provides more robust gait analysis capabilities in home environments. Emerging solutions, like mmMesh, leverage mmWave sensors and deep learning models to construct 3D human meshes, but they face challenges in capturing precise foot movement, which is critical for analyzing abnormal gaits. Existing models trained on data from healthy individuals may not adequately address the needs of patients with severe impairments, such as asymmetrical or irregular limb movements, underscoring the need for further research and development.
Invention Description:
This invention is a method to measure characteristics about a subject’s gait, including speed, length, cadence, etc. processes raw mmWave signals into dynamic Range-Azimuth (RA) Maps and FFT data and leverages a deep learning model to extract valuable features for precise foot location prediction of targets. To efficiently extract features from the time-series dynamic RAMaps, which distinguish the moving objects from surroundings, and FFT data, which preserves detailed mmWave reflections from surroundings, we design an innovative deep learning model, AutoGaitNet. This model employs five components: Dimension Extension module, which expands the dimension of FFT features; Global Feature Extractor module, which produces the global features and trainable weights from dynamic RAMaps; Detailed Feature Extractor module, which combines the extended FFT features with trainable weights to extract detailed foot features guided by global information; Walking Pattern Extractor module, which learns abnormal gait patterns from time-series features; and Foot Location Generator module, which finally predicts 3D foot locations. To capture the height information, our model learns abnormal walking patterns from inputs, combining 15 frames of time-series mmWave reflections as a single sample, to enhance the location accuracy in vertical dimension. The multiple frames of mmWave reflections also mitigate the problem of data sparsity by capturing more comprehensive information of the target. Additionally, if some frames fail to capture the subject, neighboring frames can compensate for the missing information, thus improving the quality of the generated results.
Potential Applications:
Physiology industry; medical industry
Advantages and Benefits:
This invention presents a cost-effective, timely, and practical method for measuring characteristics associate with the human gait. This method is also accurate and precise, with high-resolution tracking and a compact setup.