Artificial Intelligence usage for identifying automotive products

Resumo

The Computational Vision process has been presenting a huge development in the recent years. This is
occurring thanks to the development in the field of Artificial Neural Networks, specially the Convolutional Neural
Networks. These networks are capable of training to identify patterns in a large set of images, for latter identifying
these same patterns in other images. A very common architecture used nowadays, due to its high accuracy, is the
Mask R-CNN. This architecture not only classifies and identifies objects, but also realizes its segmentation pixel
by pixel. In this present work, Mask R-CNN was used for segmentation of automotive products (windshields,
headlights, tail lights, bumpers and rearview mirrors) in an aftermarket organization. In its evaluation, the
algorithm presented a significantly high mAP and accuracy – checked through a confusion matrix, even reaching
a val_loss of 2.413, demonstrating a satisfactory result for its proposed applications: a system filter for preventing
human error and a premise for future works of identifying defects in the mentioned products.