Sistema automático para el control de la calidad del calzado mediante visión artificial

Abstract

The objective of the current work was to implement an automatic system for the quality control of footwear. It pretends to provide the industrial sector with a tool to improve the process to evaluate the quality, identifying the existence of defective shoes through the detection of failures using Deep Learning and Computer Vision algorithms. To continue, the implementation of the automatic quality control system for footwear starts from the selection of the electronic components to be part of the project, considering the hardware and software requirements that help the compatibility between them. Subsequent, the prototype has a conveyor belt that is responsible for moving the footwear to the artificial vision booth. This compartment has an ultrasonic sensor that detects if there is a product inside it and sends the signal to the Arduino to stop the band for an estimated time of 20 seconds. Then, the four cameras capture photos and detect any problem, later, save the results to the database. In fact, the development of the system presents a detection about of footwear defect types such as threads, paint, and glue using the YOLOv5 model, which is trained through a process that is responsible for learning the neural network. Finally, the results are presented through evaluation parameters of the failure detection system through confusion matrices and validation of the results of the network training, obtaining an accuracy of 94.7%. In addition, regarding the quality of the footwear, there is an average detection of 83.80% of coincidences in the recognition.