Shyen

Inspiration

Security cameras can generate numerous motion alerts, but an alert alone does not explain the bigger picture. We wanted to explore how repeated activity could be organised into meaningful patterns so users can better understand which events may deserve attention. This inspired us to build SHYEN, a security-awareness platform focused on turning isolated alerts into understandable context.

What We Built

SHYEN combines a web dashboard, a FastAPI backend, a computer-vision video-analysis workflow, and an attention-classification engine. It analyses sampled video frames to detect people and vehicles, evaluates repeated events within defined time windows, and provides explanations for the resulting attention levels.

Instead of automatically labelling someone as dangerous, SHYEN highlights activity patterns for human review. Our goal is to make security information more understandable and useful.

How We Built It

We used Python and FastAPI for backend services, React and Vite for the dashboard, OpenCV and MediaPipe EfficientDet for video analysis, SQLite for local data storage, and Ollama with Qwen2.5-Coder for local AI experimentation. Git and GitHub supported version control and development.

We developed the project incrementally, testing the video detector, refining event-processing logic, benchmarking model performance, and connecting the components into a demonstrable workflow.

Challenges We Faced

One major challenge was achieving reliable person and vehicle detection while keeping processing time and memory usage low on CPU-based hardware. We tested different model configurations and evaluated their speed and detection results.

We also explored AI-generated explanations and encountered limitations with cloud-model access, leading us to experiment with local models through Ollama. Another important challenge was distinguishing repeated detections from meaningful activity patterns without making unsupported assumptions about a person's intentions.

What We Learned

Building SHYEN taught us that a useful security system needs more than object detection. It also needs reliable event handling, clear explanations, careful testing, and honest communication about its limitations. We learned to benchmark models, design threshold-based logic, integrate frontend and backend components, and validate features through repeatable tests.

What's Next

SHYEN is an ongoing project. Our next steps include improving detection reliability, strengthening automated tests, refining explanations, and working toward supported integration with real Ring security events. Our current demonstration uses prerecorded, AI-generated video rather than a live Ring camera feed.

SHYEN — turning repeated security activity into understandable context.

Built With

• amazon-ring
• artificial-intelligence
• computer-vision
• css
• efficientdet
• fastapi
• generative-ai
• git
• github
• html
• javascript
• machine-learning
• mediapipe
• object-detection
• ollama
• opencv
• python
• qwen2.5-coder
• react
• rest-api
• security
• sqlite
• video
• vite