Desarrollo de un sistema de visión artificial para detectar automóviles estacionados en lugares no permitidos

Abstract

This project shows an implementation of a model that will be the union between two classification techniques: convolutional neural networks and vector support machines applied in an artificial vision system. It works in real time in order to detect cars parked in places not allowed. This implementation focuses on controlling infringements autonomously. The convolutional neural networks, previously training and validated, are responsible for generating a map of characteristics of the input images. Then, this map goes into the classifier, the vector support machine which is created by a previous process of training and validation. This activity generates an output that determines whether or not a car is located in the area. The input images are the frames of a video. These images go into the model; which generates an output to confirm that there is a car in the frames. If the car goes into an infringement area and takes too much time in it, a capture of the infraction is made; otherwise the tracker follows the car until it leaves from the video frame.