AI-on-the-edge-device

Welcome to the AI-on-the-edge-device

Artificial intelligence based systems have been established in our every days live. Just think of speech or image recognition. Most of the systems relay on either powerful processors or a direct connection to the cloud for doing the calculations up there. With the increasing power of modern processors the AI systems are coming closer to the end user - which is usually called edge computing. Here this edge computing is brought into a practical oriented example, where a AI network is implemented on a ESP32 device so: AI on the edge.

This projects allows you to digitalize your analoge water, gas, power and other meters using cheap and easily available hardware.

All you need is an ESP32 board with a supported camera and a bit of a practical hand.

Key features

• Small and cheap device (3x4.5x2 cm³, < 10 EUR)
• camera and illumination integrated
• Web surface for administration and control
• OTA-Interface to update directly through the web interface
• API for easy integration
• Inline Image processing (feature detection, alignment, ROI extraction)
• Tensorflow Lite (TFlite) integration - including easy to use wrapper

Workflow

The device takes a photo of your meter at a defined interval. It then extracts the Regions of Interest (ROI's) out of it and runs them through an artificial inteligence. As a result, you get the digitalized value of your meter.

There are several options what to do with that value. Either send it to a MQTT broker, write it to an InfluxDb or simply provide it throug a REST API.

Impressions

AI-on-the-edge-device on a Water Meter

Web Interface (Water Meter)

AI-on-the-edge-device on a Electrical Power Meter

Setup

There is a growing wiki which provides you with a lot of information. Head there to get a start, set it up and configure it.

Download

The latest available version is available on the Releases page.

Initially you will have to flash it through an USB connection. Later an update is possible directly over the Air (OTA).

Casing

A 3d-printable housing can be found here:

• https://www.thingiverse.com/thing:4573481 (Water Meter)
• https://www.thingiverse.com/thing:5028229 (Power Meter)
• https://www.thingiverse.com/thing:4571627 (ESP32-Cam housing only)

Donate

If you would like to support the developer with a cup of coffee you can do that via Paypal.

If you have any technical topics, you can file a issue in this repository.

In other cases you can contact the developer via email:

Changelog

General remark: Besides the file firmware.bin, typically the content of /html will need to be updated!

---

Rolling (2022-09-18)

• Updated PostProcessing with respect to CheckDigitIncreaseConsistency

11.3.1 - Intermediate Digits (2022-09-17)

• ATTENTION:

  • first update the 'firmware.bin' and ensure that the new version is running

  • Only afterwards update the 'html.zip'

  • Otherwise the downwards compatibility of the new counter clockwise feature is not given and you end in a reboot loop, that needs manual flashing!

• NEW v11.3.1: corrected corupted asset firmware.bin

• Increased precision (more than 6-7 digits)

• Implements Counter Clockwise Analog Pointers

• Improved post processing algorithm

• Debugging: intensive use of testcases

• MQTT: improved handling, extended logging, automated reconnect

• HTML: Backup Option for Configuration

• HTML: Improved Reboot

• HTML: Update WebUI (Reboot, Infos, CPU Temp, RSSI)

• This version is largely also based on the work of caco3, adellafave, haverland, stefanbode, PLCHome

11.2.0 - Intermediate Digits (2022-08-28)

• Updated Tensorflow / TFlite to newest tflite (version as of 2022-07-27)
• Updated analog neural network file (ana-cont_11.3.0_s2.tflite - default, ana-class100_0120_s1_q.tflite)

• Updated digital neural network file (dig-cont_0570_s3.tflite - default, dig-class100_0120_s2_q.tflite)

• Added automated filtering of tflite-file in the graphical configuration (thanks to @caco3)

• Updated consistency algorithm & test cases

• HTML: added favicon and system name, Improved reboot dialog (thanks to @caco3)

11.1.1 - Intermediate Digits (2022-08-22)

• New and improved consistency check (especially with analog and digital counters mixed)
• Bug Fix: digital counter algorithm

11.0.1 - Intermediate Digits (2022-08-18)

• NEW v11.0.1: Bug Fix InfluxDB configuration (only update of html.zip necessary)

• Implementation of new CNN types to detect intermediate values of digits with rolling numbers

  • By default the old algo (0, 1, ..., 9, "N") is active (due to the limited types of digits trained so far)
  • Activation can be done by selection a tflite file with the new trained model in the 'config.ini'
  • Details can be found in the wiki (different types, trained image types, naming convention)

• Updated neural network files (and adaption to new naming convention)

• Published a tool to download and combine log files - **Thanks to **

  • Files see '/tools/logfile-tool', How-to see wiki

• Bug Fix: InfluxDB enabling in grahic configuration

Tools

• Logfile downloader and combiner (Thx to reserve85)
  • Files see '/tools/logfile-tool', How-to see wiki

Additional Ideas

There are some ideas and feature requests which are not followed currently - mainly due to capacity reasons on side of the developer. They are collected here: FeatureRequest.md

---

History

10.6.2 - Stability Increase (2022-07-24)

9.2.0 - External Illumination (2021-12-02)

8.5.0 Multi Meter Support (2021-10-07)

7.1.2 MQTT-Update - (2021-06-17)

6.7.2 Image Processing in Memory - (2021-05-01)

5.0.0 Setup Modus - (2020-12-06)

4.1.1 Configuration editor - (2020-12-02)

4.0.0 Tflite Core - (2020-11-15)

3.1.0 MQTT-Client - (2020-10-26)

2.2.1 Version Control - (2020-09-27)

2.1.0 Decimal Shift, Chrome & Edge - (2020-09-25)

2.0.0 Layout update - (2020-09-12)

1.1.3 Initial Version - (2020-09-09)

Full Changelog