CDG Hackathon 2026: Night Watch Wins with Satellite AI
Lumos team from KMITL wins CDG Hackathon 2026 with Night Watch, using VIIRS satellite imagery and AI to detect broken streetlights with 90.1% accuracy.
Stock photo for illustration only, not from the actual event
- KMITL student team wins CDG Hackathon 2026 with the Night Watch system
- Detects broken streetlights using VIIRS satellite imagery and GIS data analysis
- Reduces field inspection frequency from 33 times to just 2.7 times per month
- First-year service cost stands at 280,000 Baht, accounting for 0.49% of the lighting budget
Malfunctioning streetlights present a recurring logistical and financial burden for state agencies, requiring constant maintenance budgets and manual patrols. Highway districts typically handle streetlamp repairs around 396 times per month, yet only 12% of issues originate from citizen reports. The remaining 88 to 93 percent require staff to patrol routes manually, proving highly time-consuming and difficult to cover comprehensively every single night.
This infrastructural challenge was tackled by team Lumos from the Department of Telecommunication Engineering, Space and Geospatial Engineering program, Faculty of Engineering, King Mongkut's Institute of Technology Ladkrabang (KMITL). They secured first place at the CDG Hackathon 2026 held on September 4, operating under the AI for a Better Society Platform theme hosted by CDG Group and 18 partner universities, drawing 108 initial teams before narrowing down to 21 finalists.
Stock photo for illustration only, not from the actual event
The winning project, Night Watch, serves as a satellite-based screening system for highway lighting malfunctions. The system pulls VIIRS (VNP46A2) nighttime satellite imagery captured over Thailand every night at approximately 1:30 AM, integrating it with GIS data including a 7,695-pole asset registry, route networks with kilometer markers, traffic volume, and accident statistics.
The AI model performs three primary tasks: learning lighting patterns per route to isolate genuine streetlamps, filtering out environmental interference such as clouds and rainfall, and translating satellite coordinates into specific highway numbers and kilometer pins on a map. Testing across a 100-kilometer stretch for one month demonstrated a 90.1% accuracy rate with zero false alarms, successfully lowering inspection dispatches from 33 times monthly down to roughly 2.7.
Applying macro-level satellite imagery to local infrastructure management demonstrates a cost-effective approach to public sector modernization. It bypasses the prohibitive expenses of installing physical sensors on every individual light pole or relying on incomplete CCTV coverage. Night Watch highlights how open data and geospatial intelligence can address civic operational bottlenecks without necessitating massive physical overhaul budgets.
Additionally, the hackathon featured runner-up ClearBid AI from Chulalongkorn University, which developed an intelligent copilot for drafting government Terms of Reference (TOR) documents using RAG, OCR, and LLM technologies. The tool saves officers one full day of work per TOR document while saving agencies approximately 1.5 million Baht annually, underlining the event's core focus on practical, budget-saving technological innovations.
Source: Techsauce
Found something wrong in this article? Report an issue with this article
Comments
Leave a Comment