Análise comparativa de algoritmos de visão computacional aplicados a contagens volumétricas em área urbana

Abstract

The accelerated urban growth and the dynamism of transport systems make necessary the search for new traffic management tools, which are characterized by their capacity of processing information at a high speed, helping on the decision process. Among these tools is the use of image processing and computer vision algorithms applied to highway monitoring cameras, aiming at the obtainment of traffic parameters based on video images. The present work compares the performance of two different vehicle detection and tracking strategies applied to an urban scenario, characterized by interrupted flow. In order to do so, 45 videos were collected and analyzed and then grouped in high, medium and low flow. In terms of volumetric counts, the results indicated a superior performance of the feature based strategy, which had an average counting error of 7,41 %, while the region based strategy presented a counting error of 11,03 % on average. In addition, the feature based method reached a 74,1 % correct detection rate, whereas the correct detection rate of the region-based method was 66,6 %. The feature based strategy showed a greater capacity to handle the existing difficulties on the urban traffic, having fewer errors in terms of detection, grouping and counting.