Implementación de un sistema prototipo para la detección y recolección de botellas PET utilizando visión artificial basada en redes neuronales.

Abstract

The present graduate research project implemented a mobile device capable of detecting and collecting polyethylene terephthalate (PET) bottles using artificial vision based on neural networks. This system consists of two modules connected wirelessly via a network WIFI. A module called the collection module is responsible for the acquisition of images and the execution of movements of the mobile system, and the other module called the processing module is in charge of both the processing of images through the network neural and the control of the collecting module. The collection module was manufactured based on the 3D design of its chassis and the electronic connection of the corresponding devices to this module, while the processing module is made up only of a personal computer. In the tests, it was possible to show that the neural detection network does not present faults and can be executed even in night environments without lighting, in addition, that its delay time is only 0.019 milliseconds. Both modules can set a stable wireless connection that easily exceeds 15 meters distance with an average latency of 13 milliseconds. On the other hand, the collection system is not exempt from failures; however, with 96.7% accuracy, it is highly reliable. It is concluded that the system can detect and collect PET bottles, constituting a prototype cable to contribute to the solution of problems related to recycling. It is recommended to carry out global positioning and energy efficiency studies for complete autonomy of the device, as well as it is recommended to improve the collection system of the device.