Crime Detection and Reporting

• 

  detection

Inspiration

Prevalence of minor and major criminal scenario which could have been prevented if the controlling body were alerted beforehand

What it does

The system works to prevent any criminal activities which comprises the use of weapons including knives, guns and as such, by informing the local controlling body with a snapshot of a probable personnel with any types of weapons in the public places

How we built it

We created a model with the available image dataset from various internet sources, which were annotated for different classes to be detected, and finally the model was trained in YOLOv8 architecture, which was imported in out system framework using an API key as provided by Roboflow. The system we built continuously monitors through the use of any external camera, and when detects any anomaly in the environment including weapons or humans handling weapons, then sends a notification with the snapshot to the a next device through a portal.

Challenges we ran into

The challenges include the lack of original dataset for the training of the system. The data we were looking for could not be found in any of the platforms, as this type of system does not appear to have been built or used in any of the security system. We were imposed with resource constraints while in the process of training and building the model, as we had to train the whole model whether in our system with a single processing GPU or google colab with limited resources.

Accomplishments that we're proud of

With the vigorous brainstorming and continuous team efforts, we have been able to create a system that could aid the government and security departments to to monitor and control the probable criminal activities, without the knowledge to the public or probable victims, because the motto of our project has been to aid in a secure environment without causing any type of hinderances for the public to carry on their activities.

What we learned

During the accomplishment of the project, we have encountered various situations on how to increase the model accuracy, by creating an efficient data model with how to properly create annotation boxes, that'd help the model to properly create an efficient pattern to recognize any of such anomalies in the data to be processed.

What's next for Crime Detection and Reporting

The system we have created works finely on two of the weapons to be detected, along with the humans. But in the real world scenario, there may be other weapons not trained in the model, so these multiple weapon detection could be added in the system, along with the human detection and count to measure the vulnerability of the scenario. This model completely works on human-weapon holding pattern, so, we want to add pose detection and facial recognition to the system, for better identification along with the intent.

Built With

• javascript
• python
• roboflow
• yolov8