Evaluación de la intensidad, densidad y velocidad vehicular en la vía a través de técnicas de visión artificial

Abstract

Abstract:This document describes the application of various experiments to detection models of objects based on artificial vision techniques and convolutional neural networks, the which are YOLO v7, YOLO v5, YOLO v4, MobileNet SSD, Efficient D0 and MobileNet SSD LITE. The experiments were carried out at different times of the day with the purpose of determining the efficiency and effectiveness of the algorithms when detecting vehicles in environments with different natural lighting conditions in order to know the best detection model to carry out the evaluation of the main variables of the vehicular flow which are intensity, density and speed. In addition, StrongSort is used as the main object tracking algorithm and the use of Tensorflow, Darknet and Pytorch as DeepLearning frameworks. Finally using the Python programming language and OpenCV as the main library, developed a prototype called TraffiSmart, which has the ability to carry out the traffic evaluation through the calculation of the three variables mentioned, both in time real as in pre-recorded videos.