Abstract
The objective of this paper is to provide an analysis on detection and recognition of human motion. We focused on view-based spatio-temporal template matching. We have used background subtraction and temporal differencing by taking the required videos to extract the features and detect motion. Then Motion History Image (MHI)-based motion template is exploited to get the sequence of videos with dominant motion information. We present a Split-frame MHI concept in this paper. Histogram of Oriented Gradients (HOG) is used to describe the feature, which is extracted from an MHI template. These descriptors are trained and tested with Support Vector Machine (SVM) classifier. We have developed a new dataset of single human action at indoor environment to be named as AAMRZ. The accuracy of the strategy with well-known KTH action dataset of 6 classes is 86.6% and with AAMRZ motion dataset of 115 classes is 88%. Both the results are satisfactory based on the complexities of datasets.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have