Using IEEE DataPort for Affordable and Stable Research Data Storage

Fusion Mechanisms for Human Activity Recognition Using Automated Machine Learning

By: Ana-Cosmina Popescu, University Politehnica of Bucharest, Romania

During my research I proposed a system for recognizing human activities based on merging information from all channels of a 3D video including RGB and depth data and skeleton and context objects. These data streams are independently passed through 2D convolutional neural networks that were designed with an automated machine learning method based on the neural architecture search (NAS) technique.

The outputs of all networks are combined in a summarizing array of class scores using fusion mechanisms that are not just computationally intensive but reflect the meaningful information from a video. Additionally, another part of my research consists of filming a new RGB-D human activity recognition dataset with the purpose of making it available for the computer vision research community.

Benefits of Using the IEEE DataPort Platform
IEEE DataPort helped my research in several ways. First, it offered an affordable and stable storage method for my data. Second, it provided visibility for my research since the dataset uploaded to IEEE DataPort is much easier to find compared to uploading it on my own servers for example. Third, I could use IEEE DataPort to browse through other existing activity recognition datasets. Overall, IEEE DataPort is easy to use and provides all the necessary features for uploading a scientific dataset.

Ana-Cosmina Popescu’s research data won second place in the Spring 2019 IEEE Data Competition.