This is one of my research ideas, inspired by the existing AWLR (Automatic Water Level Recorder), which depends on sensor readings using ultrasonic, pressure, or other sensing methods. My proposed idea relies on the oldest method of reading water levels using a peilscale, but replaces manual reading with computer vision. And the most challenging aspect here is the power itself, since the last-mile device is installed in a remote area, so it will mostly rely on solar power. So, selecting the edge AI device will be the most important thing to succeed in this system.
Research Question
Under Indonesia’s Law No. 17 of 2019, River Basins are defined as integrated ecosystems channeling rainwater, with their sustainable management mandated by Government Regulation No. 37 of 2012 (BPK, 2012). Critical DAS conditions exacerbate floods and influence infrastructure decisions, yet water level monitoring remains hampered by manual peilscale errors, infrequent data, and unreliable automated tools like Valeport dataloggers and AWLR ‘BTR’ (Dhiaksa et al., 2023; Romorajausia et al., 2023). This research proposes an IoT-based AWLR that automates peilscale readings via a custom computer vision algorithm, operates on a solar-powered ESP32-S3 edge device, and enhances reliability and durability for extreme River Basins environments. The research question is: How can an IoT-based AWLR with a novel computer vision approach and solar-powered ESP32-S3 edge device improve the accuracy, reliability, and durability of water level monitoring in Indonesia’s remote River Basins? This addresses national needs for real-time, valid data to support sustainable water management.
Research Objectives
- To develop an IoT-based AWLR with a custom computer vision algorithm for automated peilscale readings on the low-power microcontroller ESP32-S3, overcoming durability, manual errors and environmental distortions.
- To optimize the low-resources microcontroller ESP32-S3 as a solar-powered edge device for local processing and IoT connectivity in off-grid river basins settings.
- To enhance reliability and durability for extreme tropical conditions (monsoons, peatlands, etc.) through robust hardware and fault-tolerant software.
- To validate performance against manual readings, AWLR ‘BTR’, and Valeport-meter in Indonesian river basins field tests.
Literature Review
Traditional AWLRs (e.g., ultrasonic sensors, pressure transmitters) degrade in harsh environments (Muste et al., 2018), while manual peilscale readings, common in Indonesia, lack real-time capability (WMO, 2010). IoT enables remote monitoring (Li et al., 2020), but power and connectivity limit its use in River Basins areas. Computer vision automates gauge readings (Siegel et al., 2019), with jomjol’s AI-on-the-edge-device (jomjol, 2023) showcasing ESP32-S3 feasibility for digit recognition. The ESP32-S3’s low-power design suits edge computing (Espressif Systems, 2023), yet its use for peilscale reading in extreme conditions is unexplored. In Indonesia, Valeport dataloggers are costly and local-storage-bound, while AWLR ‘BTR’ (NodeMCU-based) suffers from inaccurate sensors and frequent restarts (Dhiaksa et al., 2023; Romorajausia et al., 2023).
This research advances AI on the edge-device by developing a custom algorithm for peilscale-specific challenges (water reflections, day and night lighting), and integrating solar power—novel extensions addressing Indonesia’s River Basins management gaps per UU No. 17/2019.
Methodology
The study follows a multi-phase approach tailored to river basins need:
- Phase 1: System Design and Prototyping
- Hardware: Integrate an OV2640 camera, sensors (ultrasound waterlevel, temperature, humidity), and a solar system (IP67 panel, bike’s battery, solar charger controller) with the ESP32-S3, ensuring portability.
- Software: Develop a custom computer vision algorithm combining edge-canny detection and a lightweight CNN (MobileNet-based) or Tensorflow Lite, optimized for 520 KB SRAM) to handle peilscale distortions (reflections, lighting), building on AI-on-the-edge-device’s framework.
- Phase 2: Edge Computing and IoT Integration
- Process water level data locally on the ESP32-S3’s dual-core processor, using Wi-Fi for real-time IoT transmission (MQTT/HTTP) to a cloud server.
- Optimize power with deep-sleep mode and solar harvesting, ensuring autonomy in off-grid river basin areas.
- Phase 3: Reliability and Durability Enhancement
- Design for extreme conditions (monsoons, peatland flooding) with IP67 enclosures, UV-resistant materials, and anti-fog camera coatings.
- Implement fault-tolerant software (e.g., watchdog timers, adaptive thresholding for low light) to ensure stability without restarts, surpassing AWLR ‘BTR’.
- Phase 4: Field Testing and Validation
- Deploy in an Indonesian river basins (a monsoon-affected river or peatland canal) for 6-12 months, testing under rain, low light, and high humidity.
- Benchmark accuracy (MAE), reliability (uptime), and durability (failure rate) against manual peilscale readings, AWLR ‘BTR’, and Valeport dataloggers, quantifying improvements.
- Phase 5: Data Analysis
- Comparing the accuracy of peilscale readings using (1) the CNN (MobileNet SSD) algorithm and (2) the Tensorflow Lite algorithm, with peilscale pure capture from the camera as manual/human reading. These data are compared using the Pearson correlation coefficient to find their correlation. Supporting data such as energy consumption (durability), standard deviation of water level (reliability), data transmission errors (reliability), and environmental temperature/humidity are also recorded. The final result will be concluded as the best computer vision algorithm with the lowest energy consumption that can be used as the last-mile AWLR device in extreme river basins environments. The device is expected to provide water resource data (water level representation) to the Indonesian Swamp Engineering Agency in full (full for one year with 5-minute intervals), valid and accurate to support the Indonesian government’s water resource policy.
This research pioneers a novel IoT-based AWLR by integrating a custom computer vision algorithm, and solar-powered ESP32-S3 operation tailored for Indonesia’s river basin ecosystems. Unlike manual peilscale methods (error-prone), Valeport-datalogger (costly, static), or AWLR ‘BTR’ (unreliable), it delivers periodic, valid, real-time, and reliable data in a portable, durable design, aligning with UU No. 17/2019 and PP No. 37/2012. The hybrid vision model combining edge-detection and CNN/Tensorflow Lite algorithms, addresses peilscale-specific challenges (water distortions, lighting), pushing edge computing boundaries on resource-constrained hardware beyond AI on the edge-device. Its field-tested adaptability to extreme tropical conditions (monsoons, peatlands) offers new analytical potential, distinguishing it from generic tools.
Any further information on recent research updates on this topic, can contact me at hansapw@gmail.com
a Technopreneur – writer – Enthusiastic about learning AI, IoT, Robotics, Raspberry Pi, Arduino, ESP8266, Delphi, Python, Javascript, PHP, etc. Founder of startup Indomaker.com