Hi DEV community,
We are currently architecting an edge computing pipeline for a multi-zone perimeter intrusion detection system deployed across an industrial perimeter facility.
The Problem
The physical boundary relies on an array of fence-mounted accelerometer/vibration sensors and buried acoustic cables reporting to distributed edge micro-controllers (ESP32 / ARM Cortex nodes running FreeRTOS). These nodes push telemetry to a centralized on-prem Linux edge gateway over an isolated RS-485 / Modbus bus and MQTT broker (Mosquitto).
During high environmental noise (gusty winds, heavy rain, or nearby freight activity), several sensor zones begin "flapping"—firing rapid threshold transition states within milliseconds:
json
{
"zone_id": "zone-north-04",
"sensor_type": "piezo_vibration",
"trigger_val": 842,
"threshold": 750,
"timestamp": 1716982410214,
"state": "ALERT"
}
This burst triggers a telemetry storm that saturates our local broker queue and causes processing latency spikes when correlating sensor data with PTZ video verification streams.
What We've Tried
Sliding-window debounce: Implemented a leaky-bucket counter on the edge gateway (Python / AsyncIO service), requiring 3 sustained triggers within 1.5 seconds before publishing an intrusion event.
QoS adjustment: Lowered sensor-level MQTT messages to QoS 0 while keeping confirmed zone alarms at QoS 1.
Local FFT filtering: Testing basic on-node bandpass filtering to discard low-frequency wind sway before the packet leaves the micro-controller.
My Questions
How do you handle debouncing and event deduplication at the edge when raw sensor data is noisy but dropped real-world intrusions cannot be tolerated?
Would you recommend moving the sensor fusion logic entirely to edge microcontrollers (TinyML / embedded DSP), or handling aggregation centrally on the gateway via a lightweight time-series stream processor (like eKuiper or Faust)?
Any insights from folks running industrial IoT or physical security telemetry pipelines would be greatly appreciated!
Top comments (0)