Detecção e rastreamento de ciclistas em vias públicas por meio de técnicas de visão computacional

Abstract

The use of bicycles as a means of transport has attracted the attention of cities searching to provide ways for cyclists to travel safely in exchange for the use of a healthy means of transport and a reduction in the emission of pollutants produced by motor vehicles. For this, we developed a solution with the objective of performing a quantitative analysis of the cyclists using the public roads recorded in a video, as well as a discussion about the obstacles encountered in the tracking and detection of these cyclists. Detecting and tracking cyclists is done by the difference of frames, tracking of corners and motion vectors to search matching of related components found between the frames. The classification is performed by machine learning (SVM), which was previously trained with images extracted from the videos, using a modified version of the SIFT algorithm as descriptor of these images. The results obtained were medium in relation to the counting and description of the cyclists and the difficulties encountered were exposed in the evaluation of the prototype.