Towards a Simulation Platform for Generation of Synthetic Videos for Human Activity Recognition

The field of human activity recognition from video data has recently made great strides. However, the large amount of labelled data needed to train activity recognition models remains a common bottleneck. This paper introduces a simulation platform to procedurally generate synthetic videos of household activities, which randomizes portions of the virtual scene like camera position, human model, and interaction motion to introduce video variation. We describe our system design, methodology, and planned next steps for evaluation.