Principal Component Analysis Based Paralytic Attack Detection Using a New Distance Measure

Abstract Older people living alone in the home should be monitored constantly to avoid the consequences of diseases like paralytic attack. It’s very difficult to do the continuous monitoring of elder people by a caretaker manually. This motivated us to propose an automatic paralytic attack detection scheme using surveillance video analysis. Due to the unavailability of standard video dataset, we collected the training video data from different people that contain the persons’ behavior during normal conditions and in the presence of a paralytic attack (simulated). Principal component analysis (PCA) has been used to extract the features from the collected training video set and labeled it into two classes, say normal or abnormal. Basically, we analyzed the facial expression change as the criteria, and for better classification accuracy a new weighted euclidean distance based measure is introduced in this article while dealing with PCA features. An experimental study has been carried out on the collected video data, and we analyzed classification accuracy, sensitivity, and specificity. Classification accuracy obtained while using proposed weighted euclidean distance methods are compared with the other well-known classification techniques such as minimum euclidean distance based classification, support vector machine (SVM) classification, k-nearest neighborhood (k-NN). The comparative study shows that proposed weighted euclidean distance along with PCA features is able to detect faster than other methods with almost the same detection accuracy.