E-Bike Charging Station Visual Fire Monitor - reCamera 2002 64GB - fukun wu - CodeCraft Hub
Devices / BOM: reCamera 2002 64GB
This application uses Seeed reCamera and SenseCraft AI to monitor e-bike charging stations with a pure vision-based approach. The model detects smoke and flame from camera images, while Node-RED handles alarm rules, confidence thresholds, duration confirmation
Documentation
This application, based on Seeed reCamera and SenseCraft AI, targets electric vehicle charging station scenarios in parks, communities, factories, property management, and other settings, achieving pure visual fire monitoring. The system identifies smoke and fire through camera footage, completes alarm judgment and duration confirmation at the reCamera edge, and notifies alarms via MQTT and request methods.
It is designed for campus, industrial park, community, and property-managed charging areas where early visual warning is valuable. The application does not require smoke, temperature, or electrical sensors for detection input. It focuses on early-stage visual signs such as smoke and flame near e-bike battery and charging socket areas.
Note: The above solution employs a visual approach, so it is most effective in a well-lit environment
Deploy steps
- Train normal/smoke/flame classification models in SenseCraft AI.
1.Take actual pictures based on the usage scenario, and use AI to render images of smoke and fire. Each dataset contains approximately 30 images.
AI prompt:
A small amount of grayish-white smoke is generated near the electric vehicle battery, which is translucent and diffuses upwards. The original surveillance camera footage style is retained, without altering the structure of the vehicle, socket, ground, and background. Generate a small orange-red flame in the battery area of the electric vehicle, with the flame close to the bottom of the vehicle body, from the perspective of the surveillance camera, in a real and normal light environment, without altering the original scene layout.
2.On Sensecarft AI, create three categories: normal, fire, and smoke. Select the target device as recamera, and click on "Model Training"
- Deploy the model to reCamera.
1.Connect the computer and device via USB for convenient model programming
2.Deploy the model you just trained to the recamera device (if you haven't trained a model, you can directly use the model I trained below for deployment testing)
Deploy an AI model to your device (in-app tool on CodeCraft Hub).
- reCamera Node-RED editing alarm threshold and alarm method
1.Connect the device's network cable, or choose to connect the device to the network using USB
2.Enter the recamera Node-RED, add thresholds for flame and smoke detection, and configure notification methods
My threshold logic has been changed to
From normal/suspicious entry into L2, smoke is generated -> once
From normal/suspicious entry into L3, a fire occurs -> send once
"Always maintain L2 -> do not repeat"
"Always maintain L3 -> do not repeat sending"
Adjust the smoke threshold to 45, and trigger it for a duration of 2 seconds
Adjust the flame threshold to 35, triggering for 2 seconds
My notification method is to directly configure notifications to the Feishu robot (you must configure your own notification method to receive alert notifications). If you don't have a better approach, you can choose my solution, but in the notification method, you must replace the link to the Feishu robot.
- Start using the plan
1.After successfully receiving alerts using abnormal photos during training, it indicates that the project is ready for normal use and can be migrated to real-world scenarios