Devpost: https://devpost.com/software/sentinelq

SentinelQ is an edge-first security system designed to reduce false alarms and unnecessary cloud processing.

Instead of analyzing everything constantly, a low-power Arduino-based device monitors activity and only sends data to the cloud when something seems suspicious. The cloud system then runs object detection and evaluates how reliable the result is before deciding whether the situation needs attention.

The goal is simple: smarter alerts, lower cost, and more practical real-world monitoring.

• Continuously monitors the environment using low-power sensing
• Makes a first-pass decision on whether activity is suspicious
• Only escalates events when needed

This keeps the system efficient and avoids constant cloud usage.

When an event is triggered:

• A camera frame is sent to the cloud
• A YOLO-based detection pipeline processes the image
• The system focuses only on:
  • People
  • Animals
  • Vehicles

Irrelevant or low-confidence detections are ignored.

Before trusting a detection, the system evaluates:

• Brightness (to detect low-light conditions)
• Sharpness (to detect blur)
• Detection confidence

These factors are combined to estimate how reliable the detection actually is.

• Each detection is assigned a threat score
• The system decides whether to:
  • Handle locally
  • Request further verification
  • Escalate immediately

This prevents overreacting to weak or uncertain detections.

Each event is logged as a structured report that includes:

• Risk score
• Confidence level
• Short summary
• Image snapshot with bounding boxes

To avoid spam:

• A cooldown period is enforced between alerts
• Only meaningful events are reported

Most security systems either:

• Run constant cloud analysis (expensive), or
• Trigger too many false alarms

SentinelQ balances both by:

• Filtering events at the edge
• Verifying them in the cloud
• Scoring how trustworthy each detection is

• Edge Device: Arduino
• Camera Module: ESP32-CAM (OV2640)
• Backend / Cloud: Python-based detection pipeline
• Model: YOLO (real-time object detection)

• Edge-triggered cloud processing
• Real-time object detection
• Detection filtering by class and confidence
• Image quality validation (brightness + sharpness)
• Threat scoring system
• Structured incident reports
• Alert cooldown system

• Live video streaming instead of single-frame analysis
• Smarter tracking across multiple frames
• Mobile notifications
• Dashboard for viewing incidents in real time
• Model fine-tuning for better accuracy