'Safe Kerala' AI camera traffic surveillance
Released: June 2023
The state of Kerala, India, switched on an AI-powered system in June 2023 for detecting traffic violations and issuing penalties as part of its 'Safe Kerala' project, which aims to reduce road accidents and traffic violations.
Built by state enterprise Keltron and operated by Kerala's Motor Vehicle Department, the system consists of 726 deep learning-equipped cameras spread across the state that automatically detect violations and issues penalities ('challans') to violators.
The launch of the system was marred by a political controversy concerning possible irregularities in the cost estimation provided by Keltron, prompting accusations of corruption against the state's chief minister and members of his family.
Keltron also found itself in the firing line. According to The Times of India, estimates for income generated by the system provided by Keltron were 'farfetched'.
While state Transport Minister Antony Raju praised the system for detecting a dramatic increase in violations and helping reduce the number of road accident deaths, it has also been dogged by complaints and accusations of poor performance.
A closer inspection of the data provided by Raju painted a different picture. Of the 357,730 violations detected from June 5-8 2023, only 80,743 could be cross-checked and 10,457 challans issued.
It has also been widely reported that the system is known to confuse a screw or bolt on a number plate as the number zero and automatically issue a challan. It also detected a motorbike speeding at 1240 kmh, which resulted in people openly questioning its accuracy and reliability. The fine was later retracted.
Cartoq reports that Keralans have complained that the AI camera system appears to prioritise issuing challans to anyone other than politicians and ministers.
Operator: Kerala Motor Vehicle Department; Keltron
Developer: Keltron; NVIDIA; Trois Infotech
Sector: Govt - municipal
Purpose: Monitor traffic violations
Technology: CCTV; Computer vision; Deep learning; Machine learning; Object recognition
Issue: Accuracy/reliability; Bias/discrimination - income; Oversight/review; Supply chain management
Published: June 2023