In modern urban environments, inefficiency is often hidden in plain sight. Most sanitation vehicles follow fixed schedules, frequently stopping at empty bins and wasting fuel, labor, and time.
CamThink, the developer-focused brand from Milesight, is dedicated to making Edge AI accessible. By leveraging the CamThink NeoEye NE301, we demonstrate how to shift from “Scheduled Cleaning” to “Demand-driven” operations. This solution prioritizes local AI inference, ultra-low power consumption, and data privacy.
To replicate this deployment, you will need:
Note: For outdoor urban deployments, the LTE Cat.1 version is highly recommended
Expert Tip: To maximize performance on the STM32N6 chipset, select an Input Size of 320 during the quantization process. This balances detection accuracy for small-to-medium objects with the NPU’s efficiency.
Integrate the device by adding a template sensor to your configuration.yaml. This allows you to track real-time bin status and battery health.
Instead of a basic template, the NE301 integrates via MQTT. You will need to configure an MQTT Sensor in Home Assistant to parse the device’s JSON payload, allowing you to track bin status and battery life in real-time.
| Feature | NE301 Edge AI Camera | Traditional Ultrasonic Sensors |
|---|---|---|
| Accuracy | High (Visual confirmation of volume) | Moderate (Prone to errors from irregular trash shapes) |
| Intelligence | Can distinguish between trash and obstacles | Only measures distance |
| Battery Life | Ultra-low power with event-based wake | Variable |
| Privacy | 100% Local Inference | – |
A: The STM32N6 with its Neural-ART™ NPU allows for complex YOLOv8-style object detection at a fraction of the power required by traditional Linux-based gateways.
A: By using the MQTT protocol and Home Assistant’s “Collective Intelligence,” you can scale the dashboard to monitor entire districts from a single command center.
A: The server in this guide is for internal use. You need to install your own AI Tool Stack on your own
