Metal Level Measurement and Compensation Based on the MiDas Model

Accurate measurement of liquid metal levels is critical to the safety and efficiency of industrial process monitoring and automation. Traditional contact-based measurement methods face significant limitations in high temperature and extreme environments, such as safety risks, inaccurate measurements due to interference with the environment, and short equipment life. To address these challenges, this study proposes an innovative non-contact measurement method using a visual depth estimation technique based on deep learning. The MiDaS model captures an image of a liquid metal surface by a monocular camera in an industrial environment. A deep neural network then estimates their depth by processing these images. By calculating the vertical distance between the liquid metal level point and a predefined reference point, real-time and accurate monitoring of the liquid level height is realized. This study not only introduces a new non-contact solution for measuring the height of liquid metal surfaces, but also enhances the potential applications of deep learning in industrial vision measurement and depth estimation.