Automatic classification of the Parkinson’s patient stiffness from a single videosequence

This work presents a system for detection and tracking humans (Parkinson disease patients in our special case) in their natural home environment observed by a static single color surveillance camera. The purposed method emloys modified TLD tracker. The modifications enable the TLD to track multiple objects of multiple different classes at a time. The TLD can newly benefit from an external object detector. The detector (Felzenszwalb’s detector in our case) allows the system to automatically initialize the tracking procedure, it helps the tracker to keep focused on the object and to terminate the tracking correctly. Resulting tracklets are automatically sorted out between individual persons and those belonging to the patient are used for assessing patient’s stiffnes. The method was implemented and tested on an extensive (8 hours) videosequence.