Automatic Detection of Zebra Crossing Violation

Abstract

This paper attempts to provide a solution to detect a particular traffic violation rule—that of stopping on a zebra crossing at a traffic signal, instead of behind it. This violation of norm has previously neither been tackled nor attempted to be resolved. The solution involves a multi layer pipelined stages of pre-processing that include binarizing the image, filtering (that is median and Laplacian filtering), and morphological processing which help to identify if a vehicle that stopped at a traffic signal is halted on the zebra crossing or before it. A template of the zebra crossing that is under investigation is stored in the system for analysis. This template is devoid of any vehicular traffic and captures only the markings of zebra crossing. At runtime, the image captured is sent to a detector which combines the analysis of correlation through three different correlation metrics and subsequently detects any violation. This approach has been tested on manually acquired 50 test data spread over two different zebra crossings on Indian roads with an accuracy of over 90 %. This approach is simple, yet efficient and robust.