📅 Day 1 – Camera Traps, Sensor Data & Detection Basics
- Context: poaching, human–wildlife conflict, bycatch detection, and why early detection matters.
- Hands-on: load a small set of camera-trap images (classes:
human, animal, empty).
- Apply a pre-trained object detection model or simple image classifier to tag images.
- Analyse outputs: count events such as “human at night”, “animal near boundary”, and “empty frame”.
- Discuss detection errors (missed detections, false alarms) and the limits of off-the-shelf models.
👉 Outcome: Basic wildlife detection pipeline + 01_camera_trap_detection_demo.ipynb.
📅 Day 2 – Simple Alert Rules & Edge Constraints
- Concept: edge vs cloud – latency, patchy connectivity, energy limits, and device constraints.
- Hands-on: build a simple rule-based alert layer over the detection outputs.
- Example rule: if
person is detected in a sensitive zone within a risky time window → raise a “high-priority alert”.
- Log alerts into a small table with
timestamp, location, alert_type, and confidence.
- Prioritisation: distinguish high, medium, and low-priority alerts for rangers and control rooms.
👉 Outcome: Minimal threat alert engine + 02_conservation_alert_logic.ipynb.
📅 Day 3 – System Design Canvas & Governance
- Ethics & governance: privacy of local communities, false positives, ranger fatigue, and safety risks.
- Hands-on: fill a system design canvas for one park/reserve (actors, data flows, devices, policies).
- Define alert routing: who receives which alerts, via which channel (SMS, app, radio), and in what format.
- Design escalation and response: expected response times, hand-offs, and log/feedback loops.
- Group sharing and refinement: compare designs and stress-test them with “what if” scenarios.
👉 Outcome: Deployable concept note for a conservation area + 03_conservation_edge_ai_system_canvas.docx.
