The problem Sentinel AI solves

This project provides an integrated safety and security solution designed to enhance public safety and streamline law enforcement tasks. Here's how it can be used and how it makes existing tasks easier and safer:

Real-Time Tracking: This system allows law enforcement agencies to track criminal activity and police cars in real-time, facilitating quick responses and strategic deployment of resources.
Building Surveillance with Pose Detection: Using pose detection models, the system can monitor buildings for unauthorized access or suspicious behavior, helping security teams maintain a secure environment.
Violence Detection: With the integration of pose detection and LSTM (Long Short-Term Memory), this system can detect violent behaviors, enabling faster response times to incidents and potentially preventing harm.
Lost Bag Detection: The system's lost bag detection model helps identify and locate unattended bags in public spaces, reducing the risk of security threats and assisting in the return of lost items.
Overall, this system can be used by law enforcement agencies, security personnel, and building managers to improve public safety, respond to incidents more efficiently, and create a safer environment for everyone. It reduces the need for manual monitoring and enhances situational awareness through automation and advanced modeling techniques.

Challenges we ran into

Building this project posed several challenges that required innovative solutions. Here are some of the specific hurdles we faced and how we overcame them:

Real-Time Data Handling: One of the key challenges was handling real-time data streams for tracking criminal activity, police cars, and security checkpoints. To overcome this, we optimized the backend to ensure smooth data flow and minimized latency. This involved using asynchronous processing and data caching to maintain performance while updating the map in real-time.
Map Customization and Marker Management: Creating custom markers and dynamically updating their positions on the map was tricky. We had to ensure that these markers reflected the latest data accurately. We addressed this by implementing a flexible structure for marker creation and allowing on-the-fly updates to marker positions and labels.
Pose Detection Calibration: The pose detection system required precise calibration to ensure accurate detection of poses and to avoid false positives or negatives. To solve this, we conducted extensive testing with a variety of scenarios and refined the detection algorithms, making them more robust to different lighting conditions and camera angles.
Violence Detection Training: Training the violence detection model with pose data was another challenge, as we needed a dataset that represented various types of violent behavior without promoting harmful stereotypes. We collaborated with experts to create a balanced dataset and used a combination of data augmentation techniques and LSTM modeling to improve accuracy.
Integration of Multiple Models: Combining different models (e.g., pose detection, LSTM, and bag tracking) into a single coherent system required careful design and integration. We tackled this by creating a modular architecture that allowed us to seamlessly plug in different models and maintain their interoperability.
Each of these challenges taught us valuable lessons in system design, data management,