Effective Object Segmentation in a Traffic Monitoring Application

The main task of traffic monitoring applications is to identify (and track) moving targets. Thresholding an image resulting from a background difference operation is a common way to detect moving pixels and represents the starting point for all of the subsequent operations. Therefore, a wrong choice for the threshold could afflict final results. The morphological operations utilized within most of the traffic monitoring systems to remove noise dramatically undergo the effect of a “wrong” thresholding. Namely, sometimes wrong thresholded images do not allow morphological operators to reconstruct the whole signal. Besides, a fine tuning of the parameters related to morphological operators is often required. The segmentation method we present in this work has been used within the traffic monitoring system we are developing. It is based on an original morphological operation which takes advantage of all the true signals which pass through a low threshold, without being heavily afflicted by the inevitable huge amount of noise. The decision of removing, or preserving, a connected component is based on whether a given object is structured according a given pattern. The operator we developed accomplishes this task in an original way. Last but not least, the simplicity required to tune the few parameters related to the morphological operator encourages its use.