Computer Vision Art Gallery

CVPR 2021: Virtual due to COVID-19

David Rokeby

Dundas Square Surveillance Etude, (2021)

This is a study for a longer work I am planning that parses the activities in a large public space captured in 4K video in various ways using computer vision techniques. The ones used in this etude are relatively straightforward: background / foreground separation, motion / stillness separation, and multi-scalar edges. These are combined in a manner I call temporal depth of field, in which the relative focus or blur of elements is determined by temporal factors, for example whether something is moving or still, or has been in a location for a long or short time. The work fades out, at the end to a set of moving dots derived from OpenPose deep learning pose detection. These tracked dots will be featured in some unusual ways in the future full length version.

Back To The Gallery