Action Classification with Locality-constrained Linear Coding

TL;DR

This paper introduces a novel action classification method using Locality-constrained Linear Coding (LLC) to effectively encode spatiotemporal features, outperforming existing algorithms on benchmark datasets.

Contribution

The paper presents a new LLC-based encoding approach for action classification that demonstrates superior accuracy over traditional sparse coding methods.

Findings

LLC outperforms Sparse Coding in encoding effectiveness.

The proposed method achieves higher accuracy than ten state-of-the-art algorithms.

Experimental results validate the superiority of LLC in action recognition tasks.

Abstract

We propose an action classification algorithm which uses Locality-constrained Linear Coding (LLC) to capture discriminative information of human body variations in each spatiotemporal subsequence of a video sequence. Our proposed method divides the input video into equally spaced overlapping spatiotemporal subsequences, each of which is decomposed into blocks and then cells. We use the Histogram of Oriented Gradient (HOG3D) feature to encode the information in each cell. We justify the use of LLC for encoding the block descriptor by demonstrating its superiority over Sparse Coding (SC). Our sequence descriptor is obtained via a logistic regression classifier with L2 regularization. We evaluate and compare our algorithm with ten state-of-the-art algorithms on five benchmark datasets. Experimental results show that, on average, our algorithm gives better accuracy than these ten algorithms.