Bengaluru traffic system’s AI accuracy questioned after false challan

A Bengaluru motorist was wrongly issued a traffic challan after an artificial-intelligence camera mistook a guitar strapped to his back for a pillion passenger, raising questions about the accuracy of automated enforcement. The incident occurred in September when the Bengaluru Traffic Police system flagged the rider for carrying a pillion without a helmet. The rider had no passenger.

A senior officer attributed the error to the AI camera, while a manual-review failure also allowed the challan to be issued. Bengaluru Traffic Police uses an Intelligent Traffic Management System to detect violations and send images to its Traffic Management Centre. It also generates challans through officer-captured images and a citizen-reporting platform called Public Eye.

The Joint Commissioner of Police (Traffic), Karthik Reddy, said the ITMS was nearly 99 per cent accurate with minimal human intervention. However, data reviewed by The Hindu showed that the system generated more than 19,000 AI-based challans daily, while about 25 cases a day were challenged. The newspaper noted that the challenge rate could not itself measure accuracy.

Another officer said accuracy could fall to about 90 per cent after manually corrected cases were considered. The police had also found that accuracy dropped to 80 per cent by the end of 2023, particularly for pillion-riding, seatbelt and signal-jumping violations. Officials said performance had improved through machine learning but insisted that human review remained necessary.

Faded zebra crossings and missing traffic signs can also lead to questionable flags. The system mainly detects helmet, seatbelt, signal, number-plate, phone-use, no-entry and triple-riding violations. The police said human oversight was needed before enforcement action.