Performance of video processing at the edge for crowd-monitoring applications

Video analytics has a key role to play in smart cities and connected community applications such as crowd counting, activity detection, event classification, traffic counting etc. Using a cloud-centric approach where data is funnelled to a central processor presents a number of key problems such as available bandwidth, real-time responsiveness and personal data privacy issues. With the development of edge computing, a new paradigm for smart data management is emerging. Raw video feeds can be pre-processed at the point of capture while integration and deeper analytics is performed in the cloud. In this paper we explore the capacity of video processing at the edge and shown that basic image processing can be achieved in near real-time on low-powered gateway devices. We have also investigated deep learning model capabilities for crowd counting in this context showing that its performance is highly dependent on the input size and rescaling video frames can optimise processing and performance. Increased edge processing resolves a number of issues in video analytics for crowd monitoring applications.