< Project Overview >
This project is working to overcome economic and infrastructure barriers relating to CCTV networks across our cities.
The ability to retrofit and repurpose existing CCTV networks with IoT and machine learning technology could provide further opportunities to understand and monitor our cities by recognising and counting features without an operator.
However, retrofitting comes with a number of machine learning and processing challenges in order to ensure best performance.
Computer vision technology, in the context of CCTV, requires machine-learning models to be trained to recognise and capture particular objects. Only when the model is trained on sufficient volume and variety of images can it generalise enough to be deployed across a variety of different locations without further training.
This project looks to address this issue by developing an ecosystem for IoT CCTV sensing through the creation of an open training library with 10,000 labelled images, alleviating the cost and availability barriers associated with existing libraries.
The project will make use of Tyne and Wear Urban Traffic Management and Control’s API that provides open access to still imagery from over 200+ cameras. Prototype solutions will access and add extensions to Newcastle City Council’s narrowband wireless network.
The second barrier is enabling real-time processing of the video streams, which ideally should be done close to the source. A number of potential solutions, such as using cloud services or retrofitting each camera with processing equipment, come with cost high cost implications. This project will look to implement a mesh network arrangement, where nearby devices connect to each other and collectively stream video to a local processing hub. This hub is then equipped with sufficient computational power to process all the incoming streams and network connectivity to upload the results.