A professional Computer Vision-based surveillance system designed for high-reliability target tracking. Utilizes YOLOv8 and ByteTrack to monitor custom Regions of Interest (ROIs), instantly detecting and alerting when critical items (like laptops, bags, or proprietary equipment) are removed from their designated zones.
Designed as an Industrial IoT (IIoT) edge node, with built-in export capabilities for single-board computers (Raspberry Pi/Jetson) and cloud telemetry broadcasting.
- State-of-the-Art Tracking: Uses YOLOv8 for detection and ByteTrack (
lapx) to persist object IDs across frames, resisting occlusions. - Custom Regions of Interest (ROIs): Draw multiple independent bounding boxes to monitor completely different objects simultaneously in a single feed.
- Smart Alert State Machine:
- Threshold Filtering: Configurable tolerance to ignore brief occlusions (e.g., someone walking past the camera).
- Returns Grace Period: Configurable recovery frames to prevent false "SECURED" states caused by temporary tracking flickers or ID re-assignments.
- Live CV Analytics Overlay: Real-time FPS metrics, active model tracker, and confidence thresholds rendered directly onto the video pipeline.
- Dynamic Model Selector: Hot-swap between YOLOv8 Nano (
n), Small (s), Medium (m), and Large (l) models on-the-fly to test speed vs. accuracy tradeoffs without restarting the application. - IIoT Ready Pipeline: Converts CV alerts into JSON telemetry payloads published over MQTT to cloud brokers (AWS IoT / ThingSpeak).
graph TD
subgraph Edge Device [Computer Vision Edge Node]
A[Video Source / Webcam] -->|Frames| B(YOLOv8 Detection)
B -->|BBoxes + Classes| C(ByteTrack Association)
C -->|Persistent IDs| D{Spatial ROI Matching}
D -->|Target Present| E[State: SECURED]
D -->|Target Missing| F{Grace Period Check}
F -->|Threshold Exceeded| G[State: ALERT]
E --> I[Statistics Manager]
G --> I
end
subgraph Actuation & Cloud [IoT Layer]
G -.->|JSON Telemetry| J[MQTT Cloud Bridge]
G -.->|MIME Multipart| K[SMTP Email Notification]
J -.-> L((AWS IoT / ThingSpeak))
end
- Clone the repository:
git clone https://github.com/AmitAK1/missing-object-surveillance.git
cd missing-object-surveillance- Install core dependencies:
pip install -r requirements.txt- Configure the Environment:
Rename
.env.exampleto.envand fill in your SMTP email credentials if you want email alerts to fire.
Launch the GUI dashboard:
python gui_app.py- Settings Panel: Adjust your Detection Confidence (removes ghost detections) and Alert Thresholds interactively.
- Select ROI: Click "Select ROI & Start", draw a box around the physical object you want to secure, and press
SPACEorENTER. - Export Data: Navigate to the "View Data" panel to export all historical alerts and FPS performance data to CSV.
Looking to deploy this on hardware? Check the iot/ directory:
iot/export_to_onnx.py: Strips PyTorch overhead and exports the model to an optimized.onnxgraph for Raspberry Pi / Jetson Nano inference.iot/mqtt_bridge.py: Publishes lightweight alert payloads to external message brokers.
Tested on standard consumer hardware (CPU inference) at 720p resolution.
- YOLOv8 Nano (n): ~9-10 FPS
- YOLOv8 Small (s): ~3-5 FPS
- Alert Latency: < 500ms from the frame the object hits the
ALERT_THRESHOLD. - State Recovery: Grace period set by
ALERT_RETURN_THRESHOLD(default: 10 frames) effectively mitigates >95% of false recoveries due to tracking jitter.
As an engineer, it's critical to acknowledge the boundaries of CV systems.
- Heavy Occlusion (ID Switching): If a watched object is highly occluded by a passing person, ByteTrack may lose the identity and re-assign a new tracking ID upon reappearance. The ROI-matching logic handles this gracefully, but an ID switch still occurs internally.
- Low Lighting Conditions: YOLOv8's feature extraction degrades in dark environments, causing confidence to drop below the
DETECTION_CONFIDENCE_THRESHOLD, which will trigger a missing object alert. - Crowded Scenes: Heavily crowded views can cause bounding box overlap noise.
Built as a Computer Vision / IIoT foundations project.