Right now all the camera footages are seen manually and multiple people watch of crime and other incidents. We as a Team, work towards reducing the crime in a smart city wherein this manual effort is automized. Now the local PD will get a message and will know which camera to look at where the actual crime is happening. We believe in preventing the further escalation of the crime through quick notice to the officials.
What it does
This programme analyzes the footage from the security cameras and then detects it or not there is any suspicious activity(physical/sexual assault or weapon detection) happening. If yes then it instantly triggers a message to the local PD with the camera name/address along with the camera location.
How we built it
The camera footage is run through the Google vision API, understand the frame and report if any crime is happening in the frame (physical/sexual assault or weapon detection) by image captioning. And then it triggers a message to the local PD via slack using python requests library.
Challenges we ran into
Dragonboard does not support 5 GHz frequency wifi and had to use mobile hotspots. Had to rebuild the model and moved to the cloud because the local machine crashed. Accessing Android files on Mac is not easy. Integration via the internet is not fast. Dynamic file allocation is not always accessible and might run into errors.
Accomplishments that we're proud of
Making a native image captioning model. Running the model on Dragon board 410 C. Assessing the footage over the internet and returning a slack message. (Doing everything over the internet and not running locally)
What we learned
Learned to Use Auto ML and GCP. Image captioning using GCP. Learnt slack integration, android mobile hotspots can offer 2.4 and 5 GHz optionally. Learnt about dynamic file allocation and Android accessibility through Unity. Learnt a lot of unity triggers (You can trigger a condionitial phone call through unity).
What's next for ConCVe
Scale it to a complete city (Step towards smart cities) and make it into a standalone one-piece camera and run it locally and deploy with a PD. More focussed towards initially deploying it in Berkeley because of extensive crimes, security camera availability and ease of deployment and once successful will deploy in other cities/university campuses like LA and UCLA and then San Fransico.