Abstract:
The project addresses the demand for design of security systems to assess the threat levels to
valuable structures such as houses and offices. Threats deemed to be more common were observed
and strategies to monitor them were constructed accordingly. This gave rise to our method of
developing two distinct modules for threat assessment. One would be the network of generic
sensors covering the monitoring of commonly occurring dangers to a site. An example would be a
smoke sensor for the detection of fire. The other is the development of a far more flexible
framework that can be empowered to cover a large variety of threats to a site. It can further be
tailored to the cater to needs of a unique site. These are the audio nodes, that detect audios from
points sensitive to possible threats. These detected audios are then to be classified as threatening
and non-threatening, further specifying the nature of the threat if they are classified as the former.
This data is also communicated to the client, notifying them when a threat is detected via an
Android Application.
A deep neural network was developed to achieve the purpose of classification. To ensure real-time
performances, a high-power GPU, the Jetson Nano was employed to run the trained model at rapid
speeds. Seamless integration of the separate working parts was achieved by a secured online
database, the Google Firebase.
