The Indian Institute of Science is working on a way to help traffic authorities book each and every offender with the use of algorithms that analyse traffic patterns and flow.
Researchers at the Video Analytics Lab of Supercomputer Education and Research Centre (SERC) of IISC have developed algorithms that can detect traffic violations from a video feed.
The algorithms have set parameters which mimic existing traffic rules, such as lane obedience, and any deviation can be automatically picked up as a violation.
The video can be fed from the existing network of cameras that supplies images to the traffic police control room where it is manually analysed for catching traffic offenders.
IISC’s system works on anomaly detection and is built on a Conditional Random Fields (CRF) framework to extract the paths of vehicular movement on a ‘frame-by-frame’ basis.
The CRF then labels the data and helps predict the pattern of traffic flow. Another algorithm called Dynamic Time Warping (DTW) then looks for similarities in the pattern of the CRF, and when there’s a deviation, a traffic offence is recorded.
Further, the researchers claim the system can already be put to use to ease Bangalore’s gruesome traffic scenario and it would apparently take the algorithm just 20 minutes to analyse the traffic pattern at said traffic junction.