Methodology for Automatic Collection of Vehicle Traffic Data by Object Tracking

Traffic monitoring is carried out both manual and mechanically, and is subject to problems of subjectivity and high costs due to human errors. This study proposes a methodology to collect vehicle traffic data (counts, speeds, etc.) on video in an automated fashion, by means of object tracking techniques, which can help to design and implement reliable and accurate software. The development of this methodology has followed the design cycle of all tracking system, namely, preprocessing, detection, tracking and quantification. The preprocessing stage attenuated the noise and increased the classification percentage by an average of 10%. The object detection algorithm with better performance was Gaussian Mixture Models with an execution time of 0.06 seconds per image and a classification percentage of 86.71%. The Computational cost of the object tracking was reduced using Template Matching with Search Window. Finally, the quantification stage got to successfully collect the vehicular traffic data on video.