
326 ◾ Cyber-Physical Systems
© 2010 Taylor & Francis Group, LLC
21.3 Abnormal Gait Capture via Geometry-Preserved
Non-Negative Matrix Factorization (NMF)
It is important to capture different abnormal gaits during rehab training. By comparing them with
a prestored gait pattern database, we will know whether or not the detected abnormal gait cor-
responds to a certain neurodisease. Our designed pyroelectric sensors (Figure 21.3) could detect
the gait of an approaching walker. We have used machine learning algorithms (such as Bayesian
patient controlled analgesia and hidden Markov model [HMM]) to recognize the normal gaits
of different walkers [22–27 ...