A learning based approach for tremor detection from videos

This work deals with a learning-based approach for detecting tremor of hands from videos. This tremor detection problem has been represented as classification of video frames as having tremor or not. Horn-Schunk optical flow algorithm has been used in conjunction with joint entropy for feature extraction from the video frames. A training-testing paradigm has been dealt with in this work. Tremor detection using this training-testing paradigm does not make use of wearable sensors. For training of video frames using the extracted features Support Vector Machine (SVM) has been used. The results of the experiment has been shown in form of confusion matrix and precision-recall graph.