Utilizing large pre-trained monocular depth models for machining setup inspection and measurement

Monocular depth estimation has many potential uses in industrial inspection, including pose estimation, conformity checks and volumetric measurements. However, producing accurate depth maps is a challenging task, especially in production of mechanical parts, where many reflective metallic surfaces are encountered. Additionally, industrial use cases often involve working with a low volume of available data, making it difficult to train custom deep learning models for depth estimation. As a solution, this paper demonstrates the viability of pre-trained monocular depth models for inspection of 5-axis machining setups. We showcase that outputs of such deep learning models can be adjusted to specific settings, based on known values of 5-axis machine coordinates. Moreover, this can be done in a way that avoids fine-tuning the network parameters using a large amount of custom data.