Monocular depth estimation enables the creation of a depth map from a single RGB image by estimating the depth of each pixel, which represents the distance of the subject from the camera. This thesis is based on a project aimed at optimizing the transport of boxes, where it was necessary to automate the determination of the protective material’s shape using machine learning and computer vision, with depth estimation being a crucial intermediate step. The thesis describes the analysis of data used for depth estimation and the challenges encountered when working with such datasets, proposing normalization through keypoint detection. Tools such as PyCharm, TensorFlow, and OpenCV are utilized. Monocular depth estimation based on deep learning was chosen as the most suitable approach. A neural network model was developed to estimate the depth map using loss functions like SSIM and L1 loss. The model evaluation showed an accuracy of 88%, with potential for improvement through transfer learning and fine-tuning of parameters. It is concluded that depth estimation is essential for developing software to optimize box transport.