Motion data classification on the basis of dynamic time warping with a cloud point distance measure

Abstract

The paper deals with the problem of classification of model free motion data. The nearest neighbors classifier which is based on comparison performed by Dynamic Time Warping transform with cloud point distance measure is proposed. The classification utilizes both specific gait features reflected by a movements of subsequent skeleton joints and anthropometric data. To validate proposed approach human gait identification challenge problem is taken into consideration. The motion capture database containing data of 30 different humans collected in Human Motion Laboratory of Polish-Japanese Academy of Information Technology is used. The achieved results are satisfactory, the obtained accuracy of human recognition exceeds 90%. What is more, the applied cloud point distance measure does not depend on calibration process of motion capture system which results in reliable validation.

Gait recognition on the basis of markerless motion tracking and DTW transform

“…On the basis of our previous experiences on gait identification carried out with respect to highly precise marker based as well as markerless motion capture data [19, 20], DTW algorithm was chosen for motion data classification.…”

“…There are numerous applications of DTW. It was used in gait identification [19, 20], gesture recognition [21], graphical symbols, handwritten characters and footwear prints recognition [22], assessment of tennis shots and selected sports activities [23, 24].…”

“…As shown in [19, 20], dynamic time warping (DTW) is a very effective algorithm for motion sequence comparison. It dynamically scales the compared time series in the time domain, which results in minimisation of the influence of local shifts between motion phases with regard to total dissimilarity.…”

“…Tracking results on the two image sequences: p1s1 (first row) and p2s1 (second row) in frames: #0,20, 40, 60, 80, 100 …”