by Monsoon rains exposed flaw in adaptive signal detection system
PUNE – City traffic officials confirmed this week that an AI-based adaptive traffic signal system installed at several major intersections required emergency recalibration after the technology consistently failed to account for large monsoon-season puddles, which the system’s vehicle-detection cameras had apparently begun interpreting as stationary obstacles rather than simply flooded road surface.
System Performed Well in Dry Conditions, Struggled Immediately With Rain
According to smart city project coordinator Anjali Kulkarni, the adaptive signal system, designed to adjust green light timing based on real-time vehicle queue length detected through overhead cameras, performed reliably during the dry months following its installation but began generating unusual timing patterns almost immediately once seasonal monsoon rains began. “The system was reading large puddles as if they were parked vehicles blocking part of the lane,” Kulkarni said. “That confused its entire queue-length calculation, and signal timing became genuinely unpredictable at several intersections during heavy rain.”
Traffic engineers eventually traced the issue to reflective water surfaces interfering with the camera-based object detection algorithm, which had been trained primarily on dry-condition footage during its initial development and testing phase. “In hindsight, testing this system through an actual monsoon season before full deployment probably should have been a requirement,” Kulkarni acknowledged. “We learned that the hard way this year.”
Drivers Reported Increasingly Erratic Signal Timing
Commuter Siddharth Joshi, who passes through one of the affected intersections daily, said he noticed the signal behaving strangely almost immediately once the rains intensified. “Some days the light would stay red far longer than it should, other days it would change almost instantly,” he said. “I did not know it was a puddle confusing a computer. I just knew something was clearly wrong with how that intersection was behaving.”
Auto-rickshaw driver Prakash Jadhav said the erratic timing created genuine confusion among drivers accustomed to a more predictable signal pattern from before the AI system’s installation. “Before, at least the old system was consistently the same every day,” he said. “This new system was unpredictable specifically when the weather made driving already stressful enough. That is a difficult combination.”
Engineers Have Retrained the System Using Monsoon Footage
Kulkarni said the technical team has since retrained the detection algorithm using footage specifically captured during heavy rainfall conditions, teaching the system to distinguish between standing water and actual stationary vehicles based on additional visual cues including surface reflectivity and the absence of typical vehicle shape outlines. “We essentially had to teach the system what a puddle looks like,” she said. “That sounds like it should have been obvious from the start, but apparently it required an entire dedicated training dataset.”
Bohiney Magazine has covered similar early-stage failures involving AI-based urban infrastructure systems encountering real-world conditions their initial training data failed to adequately represent, noting that weather-related edge cases remain a persistent and often underestimated challenge for smart city technology deployments in regions with pronounced seasonal weather variation.
Recalibrated System Has Reportedly Improved Since Retraining
Kulkarni said early testing of the retrained system during subsequent rainfall has shown meaningfully improved accuracy, with the algorithm now correctly distinguishing standing water from actual vehicle queues in the large majority of observed cases. “It is not flawless yet,” she said. “But it is dramatically better than the confusion we saw during the first heavy rains after installation.”
City Says the Episode Will Inform Future Technology Deployments
Kulkarni said the traffic department now plans to require a full seasonal testing cycle, covering both dry and monsoon conditions, before deploying any future AI-based infrastructure systems citywide. “This was a genuinely useful, if embarrassing, lesson,” she said. “Any system like this needs to prove itself across every condition this city actually experiences, not just the convenient ones.”
Drivers Say They Remain Cautiously Optimistic
Joshi said he has noticed more consistent signal behavior since the recalibration, though he admitted his trust in the system remains somewhat cautious given the earlier confusion. “It seems better now,” he said. “I will believe it fully once we get through a full monsoon season without any more puddle-related surprises.”
City Says the Lesson Extends to Other Planned Smart Systems
Kulkarni said the traffic department has already shared the puddle-detection lesson with colleagues working on other planned smart city sensor deployments across the municipality, hoping to prevent similar oversights elsewhere. “This turned into a genuinely useful case study,” she said, “for basically every camera-based system we are planning to roll out going forward.”
Jadhav said he has noticed fewer erratic changes since the update, though he continues watching the signal closely during heavy downpours out of lingering habit from the earlier confusion.
Kulkarni said the department has also started keeping a running list of other unusual seasonal edge cases worth testing proactively before any future technology rollout across the city.
Kulkarni added that the retrained system has performed consistently well through two subsequent rounds of heavy rainfall, giving the department growing confidence the fix has genuinely addressed the core issue.
Jadhav said fellow drivers at his usual stand have started comparing notes about the signal’s behavior after rain, treating it almost as a shared informal experiment among themselves.
SOURCE: https://bohiney.com
For more satire and news commentary, visit ClickHole and Weekly World News.
