ArkAngel

Home Page
Map View
List View

The Problem

Last year, ABS Australia reported 454,900 people aged 15 or over who have experienced at least one form of physical assault in Australia. 44.% of these situations occurred outside the home. The use of CCTV has expanded rapidly in public spaces, however manual monitoring of cameras would still be not easily scalable. Furthermore, currently average response times for such incidents are around 10 minutes and relies heavily on having a witness or the victim reporting it him or herself.

The Solution

ArkAngel attempts to improve community safety and make the work of security personnel easier. The app scans live video feeds and detect levels of movement through a trained neural network. Wild violent movement can automatically alarm and notify authorities. The system is a cost-effective, automated real-time solution that can send help to victims much faster.

Security personnel have access to a web app that displays CCTV cameras in a given area. The interface allows for a list view and a map view to allow users to easily filter and monitor over an area and prevent coordinated gangster attacks over regions.

How It Works

ArkAngel mainly uses a trained neural network that detects violent movements. The website is built with a React frontend (with some Material UI) and a Python Flask backend. The site is hosted using NgRock, and the text notification service is provided by Twillio. We also use a bit of the Google Maps API to display the locations of the cameras.

Challenges Faced

Various issues with the frontend positioning various items.
A lot of time wasted fixing different problems as they came up.
Training the neural network and finding relevant videos as datasets.

Accomplishments that We're Proud of

Clean and functional user experience in the web app
Different responsive interfaces for users to manipulate CCTV information
Basic neural network that gets triggered on violent movement

What's Next for ArkAngel

It would be possible to also consider audio input to determine the occurrence of an assault situation by checking sound frequencies and the duration of distressing sounds. Furthermore detecting threatening language could also improve the accuracy of the project.