Inspiration

A highly unfortunate fact remains that criminals can remain free due to crimes being undetected, slow response times, and victims held at gunpoint and unable to alert the authorities, proving a need to right these wrongs and stop these crimes before they progress too far.

What it does

It is a threat detection algorithm that has enhanced security integrated within the smart surveillance systems that can detect any threat (handheld weaponry) and suspicious behaviours from afar.

  • Implementation of state of the art object detection and classification algorithms
  • Artificially Intelligent system providing enhanced security integrated within the surveillance systems
  • Smart threat and suspicious behavior detection
  • Detection support for situations like pre-initiation of mass shooting and bank robberies or street muggings
  • Removes the need for humans to keep track of security manually
  • Reduces the action time by immediately alerting the police and security authorities

How I built it

  • Azure virtual machine to train Machine Learning models and speed up the process using GPUs
  • Azure's cognitive services API for image classification and object detection
  • Python as backend
  • Bootstrap and JavaScript on frontend

Challenges we ran into

We learnt a lot about Microsoft Azure in this Hackathon. The greatest challenge was to speed up the process. Since the image were being processed in a video stream, it was a time consuming process where every frame was sent to Azure for analysis.

Accomplishments that I'm proud of

We are proud of our idea to prevent the increase of firearm related deaths, and we incorporated Machine Learning into a working prototype, along with a website.

What we learned

We learnt that Azure is not only a cloud based service, it is much more than that. We exported our machine learning models to tensorflow and created weight matrix files. This flexible feature of Azure helped us to integrate in our system into other technologies and speed up the process by orders of magnitude. Most of us were unfamiliar with ML technologies but using Azure, most of the complex functionalities are abstracted into simple to use user interface.

What's next for HawkVisionTech

We plan to extend the functionalities of our project to suspicious behavior detection by adding mode data sets to the training model. We also plan to add face capture technologies so whenever threat is detected, the faces of people are stored for further investigation. Model can be trained to detect petty theft/shoplifting at convenience stores.

Built With

  • azure-machine-learning-deep-learning-python-neural-networks-bootstrap-javascript-theano
Share this project:
×

Updates