Abstract: Laser directed energy deposition melt track height prediction is the key to intelligent control of the deposition process. Due to the strong nonlinear relationship between process parameters and deposition dimensions, the problem of real-time melt track height prediction based on process parameters, especially in unstable regions such as deposition start and stop, remains to be solved. To this end, a new framework for melt track height prediction is proposed, which combines the high real-time performance of coaxial vision technology with the nonlinear modeling advantages of Artificial Neural Networks. Firstly, orthogonal experiments were designed to collect single-track deposition data, which were utilized to construct a basic model. Subsequently, to effectively obtain the process characterization data, an online monitoring system for the coaxial melt pool was designed, and real-time melt track width data were obtained through the online monitoring system. Finally, the measured melt track width was input into the artificial neural network as one of the features. The experimental results show that the method has good in-real-time prediction accuracy with an average relative error of less than 7% and a response time of less than 20 ms, which provides a feasible solution for the online prediction of melt track height prediction by directed energy deposition.