How it works
DeepAlert performs real-time video analytics on a continuous basis, day and night, analyses what is in the scene and delivers relevant and timely alerting based on individual alerting preferences. For example, you might want to be alerted when someone is climbing over your wall at night(!), when there is an accident on the highway, when the elephants or giraffe are near the waterhole at your favourite game park, when there is a fire on the mountain or, as a scientist, when some new species of fish swims past your remotely deployed underwater camera.
DeepAlert can be applied to many domains and is carefully architected to take advantage of the latest computing advances while recognising real world constraints (cost, power, bandwidth, etc). Some examples of actual real-time alerts from the system are shown here based on the neighbourhood watch, business watch and wildlife monitoring scenarios.
For a system which can be widely and affordably deployed and yet retain powerful analytical features, careful design is needed. DeepAlert can be globally deployed, can integrate with most IP cameras, is highly cost effective and yet is also one of the most powerful systems, utilising the very latest state of the art in machine learning. These are the components:
The heart of the system is a deep learning analytics engine which is located in the cloud, but machine learning is applied across the system. DeepAlert Hub technology can be deployed at the “edge” in order to manage the video streams and perform edge analytics in order to reduce internet bandwidth requirements. Alternatively, cameras at the edge can be configured to communicate with DeepAlert directly.
The system can integrate with personal (e.g. web and mobile) and commercial alerting platforms and is designed to integrate with a wide range of IP cameras. It can thus be deployed for new camera installations or as an addition to existing installations.
The system is “smart” and, with some quality human attention applied to its inputs, improves over time.
The alert web interface has rich display and configuration capabilities including history search of alerts:
Site setup is made simple via the hub web interface which is also remotely accessible: