Desarrollo de un sistema embebido basado en FPGA y visión artificial implementado en un vehículo para detectar y alertar sobre la presencia de canes en la vía.

Abstract

According to the National Transit Agency (NTA), there is an accident rate related to animal son the roads, therefore, the present research aimed to develop an Embedded System based on FPGA and computer vision implemented in vehicles to detected and alert the presence of dogs. The methodology consisted of four stages, considering implementing a trained Artificial Neural Network (ANN) model using the transfer learning technique as a fundamental pilar. For this purpose, to enrich and train this ANN, a set of 2,000 images was collected, of which 1,500 were allocated for training and 500 for testing. The training was developed on the Google Colab platform, and the detection algorithm was developed in Jupyter Notebook, obtaining two modes of operation: normal and detector. In the validation stage of the trained ANN, the learning curve was obtained, emonstratinng impressive accuracy and minimal data los. The robustness and effectiveness of the System were subjected to field tests, conducted in controlled and uncontrolled enviroments, yielded consistent and reliable results, with the Tukey test determining optimal performance in the morning sessions. The manufacturing cost of this embedded system was established at a total of $421.55. In this context, it is concluded that the device supports its functionality, and it is recommended to use another hardware board to optimize performance, thus leaving this research as a solid foundation for future developments.