Robust Action Recognition framework using Segmented Block and Distance Mean Histogram of Gradients Approach

This paper propose two novel algorithms, segmented block of mean image with normalization and distance mean histogram of gradients for generating descriptor. Feature analysis and classification done with the help of Random forest. Our approach performs better than benchmark, gradient based approaches with average accuracy 56.59% on HMDB dataset. We have also tested our approach on ATM video dataset. Video sequences have been analyzed by varying block size of mean image and Number of frames for mean image. Average accuracy 94.5% has been achieved during testing on ATM dataset, where proposed framework has been able to recognize activities efficiently.