Contagem volumétrica de veículos em vias urbanas

Abstract

The design of urban roads must take into account, among other things, the quantity of vehicles driven on the roads, an increasingly difficult information to obtain due to the growth of the vehicles’s population. Some of the ways of getting such information include the assignment of people to do the counting or the placement of sensors on the road. The purpose of this work is to provide an automatic computational approach for counting vehicles through video processing (obtained from urban roads recordings). This work consists on applying a series of computer vision algorithms over the videos in order to segment and count the vehicles. Many processing techniques are used, each with its own objective of extrating some information or enhacing some feature of the image (video frame). The proposed approach utilizes the background subtraction technique to detect moving objects and, after applying morphological filters, each image of the video, represented by blobs at this point, is analyzed so it can be known which blobs represent cars and, if so, how many. The blob analysis is performed in two ways, one by checking its shape and area and another by tracking the speed of the points of interest of the image. The result of the project is promising. Out of the 274 vehicles in the video, the area approach misses a single one and counts 17 extras, representing a 6% total error, yet the speed approach gets it wrong 18% of the time. A calibrating phase is necessary so that the approach can work with videos that were recorded from different angles.