Quadruped robots can adapt to various complex terrains due to their special leg structures, while most of the current virtual model-based QP (quadratic planning) algorithms do not take terrain information into account or are not comprehensive enough, which limits the stability and accuracy under complex terrains. In this paper, we propose a method to estimate the terrain complexity based on the robot ontology sensing system, and improve the quadruped robot controller based on this method to enhance its stability under unstructured terrain. In this paper, the ontology perception capability, end-effectors kinematics and centre-of-mass momentum feedback of the quadruped robot are firstly utilised to design a comprehensive terrain complexity estimation function, which combines the terrain evaluation of the quadruped robot with the dynamic performance evaluation to evaluate the terrain complexity. Then, on the basis of the virtual model control of the quadruped robot, external disturbance compensation and support force constraints using the terrain complexity estimation function are added in the stance phase to improve the stability of the control algorithm in the support phase, while the estimation function is used in the swing phase for the planning of the landing point and the adjustment of gait cycle to improve the dynamic capability and adaptability of the robot. In order to verify the effectiveness of the method proposed in this paper, a series of experiments are designed using the UnitreeA1 simulation model of the quadruped robot and the webots simulation software, and the experimental results prove that the method proposed in this paper can effectively improve the stability of the quadruped robot when working on unstructured terrain.