How to meet the personalized needs of stroke patients in rehabilitation training has long been the core challenge in rehabilitation robot research. The rapid and accurate recognition of patients’ motion intentions is a critical prerequisite for addressing this challenge. To address this, this paper presents a method for estimating human lower-limb motor capability based on a human-machine coupling model. Unlike the integrated equivalent model of human-machine coupling, the dynamic models of the mechanical leg and the human lower limb, along with the contact model, are established separately. Identification experiments are then designed to determine the unknown parameters. Then, the mechanical leg and human lower limb are integrated via the contact model to establish a human-machine coupling dynamic model. Based on this model and utilizing torque and interaction force sensors on the mechanical leg, the estimation of human lower-limb motor capability is indirectly realized. Finally, experiments validate the feasibility and effectiveness of the proposed approach.