Efficient, Edge-based Video Streaming for Video Action Recognition

Streaming video applications in the home (video doorbells, security cameras, babycams, robot vacuums etc.) have started to become commonplace. Automatically detecting events of interest in such video streams can help to reduce manual effort from users in sifting through false alarms, enhance user safety, comfort and satisfaction, as well as enable new applications in areas such as safety and wellness. This project focuses on systematically investigating the performance of action recognition models under resource constraints found in practical deployment scenarios such as variable and metered network bandwidth, limited compute, etc. as well as develop technologies to overcome these challenges.

Faculty Supervisor:

Animesh Garg

Student:

Partner:

Samsung Electronics Canada

Discipline:

Computer science

Sector:

Technology; Information and Communications Technology; New and Digital Media

University:

University of Toronto

Program: