Human gait classification using microDoppler time-frequency signal representations

A time-frequency classifier is applied for human gait classification. Time-frequency (t-f) quadratic distributions are used for time-frequency signal representations. Three specific motions are considered, corresponding to three different scenarios of arm motions which describe free and confined arm swings. The microDoppler signature in the time-frequency domain of each motion style is viewed as a feature and is incorporated in a distance-based classifications measured between the test data t-f distribution and the training average t-f distributions. It is shown that the time-frequency classifier performs properly, yielding low probability of classification errors, and its performance is rather insensitive to the type of time-frequency distribution employed. Among the possible distance measures used, which include the Correlation, Bhattacharyya and Kolmogorov, the Euclidean distance provided the best results.