Study of classification techniques for human action recognition based on PCA and width vectors

The article presents the results of the research of classification techniques for human action recognition based on PCA and width vectors using a video recorded in the optical range. The method used in this paper consists of the detection of a moving person on a video sequence with size normalization, the formation of a set of subsequences, and feature vectors. To form the feature vectors, we use two different approaches. The first approach is based on the dimensionality reduction using principal component analysis, and the second one utilizes the width profile of a detected silhouette. The classification of human actions is carried out using a support vector machine with different kernels, k-nearest neighbors, and random forest classifier. The obtained results allowed us to find the most effective parameters for the considered technique.