Recursive Image Segmentation for Vehicular Traffic Analysis

Abstract

Many methods have been proposed for image segmentation in vehicular traffic analysis using traffic camera video footage. However, isolation of moving objects with perfect object boundaries has been a challenging problem in vehicular traffic analysis. Usually these vehicle objects are extracted inside rectangular boundaries with extra irrelevant background image pixels from other objects included in the analyzed image. Thus using such segmentation methods in vehicle identification using video is not favorable for feature extraction for classification of vehicle category. This work proposes a method to deal with irregular shaped image segmentation for vehicle identification using a recursive algorithm. A binary thresholded image composed of white and black pixels is filtered with a 2D low pass filter to isolate irregular shaped image boundaries of objects. Then recursive image segmentation is applied on the filtered binary image. White pixels in the 2D filtered image are used to identify the presence of the object. If the neighboring pixels of the pixel of interest are also white, then those neighboring pixels are recursively processed the same way to account for the extent of the object. This recursive collection of pixels bounded by an irregular shaped boundary is continued until neighboring pixels are significantly different in color from the pixel of interest. From this recursive image segmentation algorithm, extraction of all pixels of odd shaped objects done in an efficient manner. Accordingly, pixels count, height and the width of the object are recorded. This image segmentation method has been successfully applied to identify vehicle categories in traffic video sequences.