What is it about?
In this paper we introduce an algorithm for 3D human action recognition using feature discrimination and unsupervised learning. Unsupervised feature learning allows us to extract features from unlabelled video data, avoiding the cumbersome process of designing feature extraction by hand. We evaluate this technique using the MSRAction data set where our method, outperforms earlier methods.
Featured Image
Why is it important?
This work is important as it introduces the technique of unsupervised learning for action recognition. This allows a serious of well characterised sequences to be learned without human input and then subsequently characterised. The initial results are favourable and indicate and that the technique is applicable to data from different sources.
Perspectives
From my perspective, the key importance of the work presented here is that the algorithms can be easily modified to work with data from different sources. Unlike techniques based on supervised learning which work using hand designed features, this work avoids this cumbersome process. The key benefit is that this technique becomes applicable to action recognition from different data sources and even different forms of characterisation tasks.
Dr Daniel Stephen Clarke
Cranfield University
Read the Original
This page is a summary of: Combining unsupervised learning and discrimination for 3D action recognition, Signal Processing, May 2015, Elsevier,
DOI: 10.1016/j.sigpro.2014.08.024.
You can read the full text:
Contributors
The following have contributed to this page
